CHAPTER THREE: METHODOLOGY

Introduction

In this chapter, the methodology of this study is presented. This chapter also aims to justify the research design. The theoretical and practical concerns of using a population-based approach and the indirect sisterhood method in particular, are explored. This chapter also discusses the data collection procedure involving staffing and recruitment of data collectors, team leaders and supervisors; the training of the research team and how data collection was conducted in the field. The survey time frame was eight months, which was February - September 2005. Data collection involved the following procedures: staffing and training and the actual collection of data

3.1. Study method

The methodology for this study is survey. “A survey is a method of collecting information from people about their ideas, feelings, health, plans, beliefs, and social, educational, and financial background” (Fink and Kosecoff 1998, p: 11). To calculate maternal mortality rates from survey data, two important and innovative methods are currently being used and they are the indirect and direct sisterhood method (Stecklov 1995). The Indirect Sisterhood method was developed by Graham et al in the late 1980's (Graham et al 1989). It provides a method for calculating the indexes of maternal mortality in countries or regions where data on vital events are not routinely and reliably collected (Hanley, Hagen and Shiferaw 1996). The Direct Sisterhood method was a variant of the Indirect (Ruternberg and Sullivan 1991).

3.2. The Sisterhood Method

The sisterhood method is an indirect technique for deriving population-based estimates of maternal mortality. This method is often recommended by the WHO and UNICEF especially for countries with inadequate registration system and low income resources (WHO 1996). The sisterhood method is based on a rationale that is similar to of the "sibling-survivorship" method. The sibling-survivorship method starts from the proportion of brothers or sisters already dead among survivors currently age x to estimate the probability of dying between birth and age x. Adopting this method for maternal mortality, identified some issues: the necessity of the respondents knowing the cause of death; all maternal deaths had to have occurred between ages [Alpha] and [Beta] (the minimum and maximum ages of childbearing); the most widely used index - the Maternal Mortality Ratio (MMR) - having a fertility component (Garenne and Friedberg 1997).

The sisterhood method is based on answers to four questions:

  • How many sisters have you ever had who ever married (or who survived until age [Alpha])?
  • How many are dead?
  • How many are alive?
  • How many died while they were pregnant, during delivery, or within six weeks after delivery (that is, died of maternal causes)?

Source: Graham et al 1989

The sisterhood method of indirectly estimating the MMR was developed because of the difficulty and expense in getting such data in other ways, the relative ease of data collection, the relatively small sample size needed, and the relative ease of calculation (Graham et al 1989). This method has become an important tool in developing countries and has been widely adopted and adapted (WHO 1997c). The original sisterhood method is indirect, this is because the MMR is not measured directly but derived mathematically from information provided by respondents about their sisters (Graham et al 1989).

The second method which is an adaptation of the sisterhood method is known as the direct method (Ruternberg and Sullivan 1991). This is a direct method of estimation that unlike the indirect does not require model fertility and mortality schedules (Stecklov 1995). The direct sisterhood method has been routinely added to many surveys including the Demographic and Health Surveys (DHS) (WHO 1996). The direct method relies on the same basic principle as the indirect that is the use of information reported by siblings to expand the sample size and to compensate for the fact that the deceased are not available for interview (Graham et al 1989). However, more data are collected on the circumstances and timing of the sister's death, which allows a more direct calculation of the MMR. The direct method requires a larger sample size, a larger number of questions, and a more difficult calculation than the indirect method (Rutenberg and Sullivan 1991). However, (table 9) due to the larger sample size, the direct method can be modified to obtain an estimate for a more recent period, and the greater number of questions provides internal validity checks (WHO 1997c).

3.2.1. Justification for the indirect sisterhood method

This study was conducted using the indirect sisterhood method. Comprehensive vital registration systems able to capture a reasonable proportion of maternal deaths are only available in very few developing countries (WHO 1997c). Where the vital registration system is inadequate, population-based surveys have to be used to estimate maternal mortality.

major difficulties to using the population based approaches have been identified (WHO 1997c; LeCoeur, Pictet, M'Pele and Lallemant 1998). These include the fact that maternal mortality is a rare event. It has been identified that even in places where levels of maternal mortality are high, the actual number of maternal deaths is likely to be relatively small because maternal deaths are rarer than infant deaths for a fixed reference period (WHO 1997c). This thus implies that large sample sizes would be needed to gather reliable results. This large sample size may involve visiting 200,000 households, which is an impossibly large number in any setting and totally unrealistic in small countries (Kwast 1985; WHO 1997c; LeCoeur et al 1998). This method is also totally unrealistic for a PhD that has to be completed within a specific time frame and has limited financial resources.

To overcome this problem of large sample size, an alternative and more efficient technique was developed. The sisterhood method is one of such approach (Graham et al 1989) and it was specifically designed to overcome the problem of large sample sizes. This is done by obtaining information from respondents during interview about the survival of all their adult sisters (WHO 1997c).

According to the WHO, many countries have used the sisterhood method during the past few years and they include: Gambia (Graham et al 1989), The Republic of Djibouti (David et al 1991), Malawi (Chiphangwi et al 1992), Niger (De Groof et al 1993), Ethiopia (Shiferaw and Tessema 1993), Zimbabwe (Oosterhuis 1993), Indonesia (Wirawan and Linnan 1994), Bangladesh (Shahidullah 1995) Kenya (Mace and Sear 1996), northern Nicaragua (Danel et al 1996), Tanzania (Olsen et al. 2000; Font et al 2000; Mbaruku, Vork, Vyagusa, Mwakipiti, and Roosmalen 2003), Uganda (Orach 2000), Ghana (Smith et al 2001) and Swaziland (Lech and Zwane 2002). However, before the indirect sisterhood method can be used to measure maternal mortality a number of issues must be taken into account (Graham et al 1989; WHO 1997c; Olsen et al 2000) these are:

That the sisterhood method is only appropriate when:

  • there is no reliable estimate of the level of maternal mortality;
  • an approximate level of maternal mortality is needed for advocacy purposes and to draw attention to the problem;
  • resources do not permit any other approach for measuring maternal mortality in the immediate term; and
  • a starting point is needed for more detailed follow-up of maternal deaths identified during the recent past.

When there is no reliable estimate of the level of maternal mortality

In Nigeria, maternal mortality ratio (MMR) estimates from nationaland international sources differ substantially. MMR values rangefrom 740 to 1500 per 100 000 live births for the last decade (WHO 2003; UNFPA 2000). The 1999 Multi Indicator Cluster Survey, estimated a maternal mortality ratio (MMR) of 704 deaths per 100,000 live births for a period of six to twelve years preceding the survey (Federal Office of Statistics (FOS) and United Nations Children Fund (UNICEF) 2000). This survey found a significantly higher rural than urban MMR (828 versus 531), and considerable variance across regions, ranging from 165 in the Southwest to 1549 in the Northeast (National-Planning-Commission (NPLC) and UNICEF 2004; Federal Office of Statistics (FOS) and United Nations Children Fund(UNICEF) 2000).

An analysis of maternal death at the University of Ilorin Teaching Hospital (U.I.T.H.) Ilorin over a 12-year period (1972-1983) identified 624 deaths making a maternal mortality rate of 4.50 per 1000 births (Adetoro 1987). The MMR in Ahmadu Bello Teaching Hospital (ABUTH) Kaduna was 652 per 100 000, and there was no consistency in the yearly trend, which varied between 321 and 730 per 100,000 live births from 1990 to 2000 (Onwuhafua et al 2000). Given the lack of consistentestimates to aid health policy and district management, thisstudy reports the life time mortality risk and MMR for Ibadan North and Ido Local Government areas obtained with the sisterhood method, as well as themain causes of maternal deaths.

When an approximate level of maternal mortality is needed for advocacy purposes and to draw attention to the problem

Worldwide, Nigeria has the 47th highest gross domestic product (GDP) and remains the world's 8th largest exporter of petroleum and Sri Lanka on the other hand, is 78th on the GDP list (Weeks 2007). When the MMR of the two countries are compared, Nigeria has an MMR of 800 per 100,000 live births but Sri Lanka has an MMR of only 92 (WHO, 2004a).

Reducing maternal mortality and morbidity emerged as a global priority two decades ago, yet levels of maternal death have not appeared to have declined significantly in the developing world (Thompson 1999). Although knowledge of medical and public health interventions is substantial, and most maternal deaths are preventable (Fathalla 2007), half a million women die each year from complications of pregnancy and childbirth. Another 15 million suffer life-threatening disabilities (WHO 2004a). Maternal mortality is an enormous tragedy worldwide, hence ought to attract a lot of attention. However, political leaders are burdened with thousands of issues, multiple political pressures, and limited resources (Thompson 1999; Weeks 2007; Shiffman and Okonofua 2007). To change that, maternal mortality ought to be seen not just as a moral or a medical issue but as a political issue - one that requires good research that allows advocates to be strategic in affecting policy (Weeks 2007; Shiffman and Okonofua 2007). In 2004 at the tenth Anniversary meeting to mark the International Conference on Population Development political will was cited as the missing ingredient in global reduction of maternal deaths (International Conference on Population Development 2004).

Compared to other health concerns, maternal mortality reduction is given low priority on government health agendas and sometimes it might be missing altogether (Shiffman et al 2006; International Conference on Population Development 2004; Shiffman and Okonofua 2007). Shiffman et al (2006) identified the key factors that must be available in a country to generate political will. These range from credible evidence of a significant problem to a united network of advocates who have identified a clear set of national policy priorities. Using available data, plan for advocacy can be developed, and a sensitisation campaign organised to inform policymakers and community members about the devastating impact that not putting additional resource toward improving maternal and health care would have on the country's economy.

When a starting point is needed for more detailed follow-up of maternal deaths identified during the recent past.

According to WHO (1997c), the original indirect method is relatively simple to implement. This simplicity is highly essential for the Researcher since to the knowledge of the Researcher, this study is the first community based study on maternal mortality in Nigeria. This is the first time also that the indirect sisterhood method would be used in Nigeria to obtain a population based estimates of maternal mortality. This meant that there is no reference point for the investigator, so a simple tool will be easier for her to manage to prevent confusion.

The indirect sisterhood method also involves only four simple questions. This made training and supervision of data collectors relatively straightforward (WHO 1997c). The questions for the direct sisterhood method are more complex and time-consuming to administer (Rutenberg and Sullivan 1991). Thus requiring additional efforts directed to the training and supervision of data collectors to prevent misreporting. Furthermore, as stated earlier, this study is a PhD project which had a time frame and required the use of a less complex and a less time- consuming tool, if the project was to be concluded in the specified three years.

In addition, the indirect sisterhood method is practical to add on to comparatively small scale surveys (WHO 1997c; Graham et al 1989; Wirawan and Linnan 1994; Hanley et al 1996). Although the direct sisterhood questions can also be added to an ongoing survey it requires more time than the four questions of the original method. This may have financial implication to a resource-limited country, such as Nigeria. The direct sisterhood method requires larger sample sizes and is, therefore, more expensive to implement (WHO 1997c).

Table 9: Summary of strengths and weaknesses of original indirect and direct variants of the sisterhood method

Strengths

Limitations

Original indirect method

  • Four simple questions can be added to ongoing household survey
  • Minimal time requirements
  • Minimal sample size requirements
  • Simple calculations to estimate ratios
  • Additional information can be gathered on place/time/cause of death
  • Can be adapted for use at facility level
  • Inexpensive
  • Care needed in the use and understanding of the questions
  • Provides retrospective estimate (10-12 years prior to the survey)
  • Not appropriate for use in settings with high levels of migration
  • Not appropriate for use in settings with declining or low fertility (TFR<3)
  • Appears to underestimate adult female mortality compared with independent empirical data
  • Not appropriate for monitoring in the short term

Direct method

  • Can be added to ongoing multipurpose household survey
  • Sibling histories permit internal data quality checks
  • Smaller sample size requirements than household surveys but larger than indirect methods
  • Can be used to provide more recent estimates than the indirect method
  • No assumptions required about patterns of fertility
  • Relatively inexpensive
  • Data collection more complex and takes longer than for indirect method
  • Separate time period estimates subject to wide standard errors
  • Not appropriate for use in settings with high levels of migration
  • Not appropriate for use in settings with low fertility (TFR<3)
  • Appears to underestimate adult female mortality compared with independent empirical data
  • Not appropriate for monitoring in the short term

Source: WHO (1997c). The sisterhood Method for estimating maternal mortality: Guidance for potential users. WHO: Geneva

3.2.2. Calculating MMR in Indirect Sisterhood Method

The Maternal Mortality Ratio (MMR) is a measure of the risk of death directly related to the pregnancy once a woman has become pregnant. It is usually measured in terms of maternal deaths per 100,000 live births. The maternal mortality ratio, which is obtained by dividing the age-standardized maternal mortality rate by the age-standardized general fertility rate, is often considered a more useful measure of maternal mortality since it measures the obstetric risk associated with each live birth.

It is calculated by dividing the number of maternal deaths regardless of pregnancy outcome (ectopic pregnancy, fetal deaths and live births) by the number of live births in a defined population (WHO 1994). Maternal mortality rate on the other handrefers to the number of maternal deaths in a period (usually a year) per 100,000 women of reproductive age (usually defined as aged 15-44 or 15- 49). This indicator takes into account both the risk of becoming pregnant and the risk of dying for reasons related to maternal complications during pregnancy (WHO 1994).

The calculations underlying the measurement of the MMR by the Indirect Sisterhood Method are described by Graham et al. In the Indirect Sisterhood Method, respondents aged 15-49 years old are grouped into five year age groups and this provides the number of sisters per age group. These counts are only for those sisters who have entered the reproductive period (>=15 years). Since the number of sisters who have entered the reproductive period reported by respondents in the younger age groups will exclude those sisters who are yet to reach reproductive age; a compensation is made to arrive at the expected number. The expected number of sisters reaching reproductive age for respondents in the younger age groups (age 15-19 and age 20-24) is calculated by multiplying the number of respondents by the mean number of sisters among respondents aged 25-49. This approximation is based on the assumption that the mean number of sisters per respondent is the same in each age group of respondents.

The total number of sisters who had died irrespective of the cause for respondents in each age group is calculated as well as the total number of sisters who died from maternal causes for respondents in each age group. The number of deaths from all causes across the age groups of respondents is summed up to give an overall total number of dead sisters. Summation of the number of maternal deaths across the age groups of respondents is done to provide an overall total of maternal deaths of sisters.

The total number of women-years of exposure to maternal mortality is calculated over the entire reproductive period. This is also called the sisters units of exposure (Graham et al 1989). This is calculated for each age group of respondent by multiplying the total number of sisters of respondents reaching reproductive age by an adjustment factor. To calculate the overall total of women-years of exposure, the number of women-years of exposure to maternal mortality across the age groups of respondents is summed up.

To calculate the probability of maternal death, the total number of maternal deaths is divided by the total number of women-years of exposure to maternal mortality for each age group. This is the lifetime risk of dying from maternal causes for a sister within the age group of the respondent.

To know the proportion dying from maternal causes within the age group of respondents, the total number of maternal deaths is divided by the total number of dead sisters within each age group. This is the proportion of maternal deaths in females of reproductive age (PMDF) within the age group of the respondent.

The overall total number of maternal deaths is divided by the overall total number of women-years of exposure to maternal mortality across all age groups of respondents from 15-19 to 45-49 to give a probability Q, which is the overall lifetime risk of dying from maternal causes for a sister of a respondent. Divide the overall total number of maternal deaths by the overall total number of dead sisters to give the overall proportion of maternal deaths in females of reproductive age (PMDF).

To calculate MMR, Graham et al (1989) used the formula 1 - [(1 - Q) 1/TFR], where Q is the overall lifetime risk of dying from maternal causes and TFR is the Total Fertility Rate. When this number is multiplied by 100,000, it gives the estimated number of maternal deaths per 100,000 live births. To correct for pregnancies not ending in a live birth using the suggestion in Hanley et al (1996), the formula 1 - [(1 - Q)1/1.2TFR] is used, where Q is the overall lifetime risk of dying from maternal causes and TFR is the Total Fertility Rate.

3.2.2.1. Adjustment factors

Adjustment factors were derived by Graham et al (1989) from a theoretical consideration of age patterns of maternal mortality, based on a model describing the relationship between the proportion of sisters dead by the age of a respondent and the lifetime risk of maternal death. Graham et al (1989) give a table of adjustment factors, which are as shown in Table 10 for this study.

Table 10: Adjustment factors for estimating women-years of exposure to maternal mortality by age group of respondent

Age group of respondent

Adjustment factor

15-19

0.107

20-24

0.206

25-29

0.343

30-34

0.503

35-39

0.664

40-44

0.802

45-49

0.900

Source: Graham et al (1989)

3. 3. Study methodology

This study was divided into four main parts:

  • Identification of the total live births in the study settings to serve as denominator
  • Identifications of cases of maternal deaths based on vital statistics and hospital records
  • Identification and measurement of maternal deaths based on community survey using indirect sisterhood method; and
  • Investigations of the causes of maternal death.

3.3.1. Study design

This was a retrospective study. Information was obtained directly from individuals at the community level using structured interviews to obtain data on maternal mortality. This was to enable a comparison to be made between the primary data collected at the community level and the secondary data available at the Ministry of Health to assess the adequacy and the quality of registration of maternal death in the state.

3.3.2. The Target Population:

The target population of this survey was all adult females and males over age 15 years in the study setting. This means that the subjects for this study consisted of one female or male from each household who was between the ages of 15-49.

3.3.3. Sampling design

A multi-stage sampling with stratification and clustering was considered appropriate for this study. Sampling is the process of selecting a portion of the population to represent the entire population (Polit, Beck and Hungler 2001). A multi stage sampling is an extension of cluster sampling where a hierarchy of clusters are chosen going from larger to smaller (WHO 2004b). A cluster is a naturally occurring unit or grouping within the population. It is a unit for which the administrative level has a clear, non-overlapping boundary (Polit et al 2001). Clustering was very useful for this study as it avoids having to compile exhaustive lists of every single person in the population. Clusters should be as heterogeneous as possible within and as homogenous as possible between.

Stratification on the other hand is the process by which the population is divided into subgroup (WHO 2004b). Sampling is then conducted separately within each subgroup. Strata should be as homogenous as possible within and as heterogeneous as possible between. This means that strata should be formulated in such a way that individuals belonging to a stratum should be as similar to each other as possible with respect to key variables from individuals belonging to a different stratum. Stratification was important in this study because of the existing geographical structure of the state and the fact that registration of deaths, although inadequate, is available in the state Ministry of Health, which is located in the state capital, to enable an assessment of the system of registration.

As earlier stated the sampling design for this study is a multi-stage sampling with stratification and clustering. Although the sampling unit is the individual subject, the sampling was conducted on the household level. The subjects were chosen by selecting a household and one eligible subject in the household was included in the sample.

Applying the multi stage stratified cluster sampling

  • Strata: Provinces (Ibadan)
  • Primary Sampling Unit (PSU): Counties (Ibadan North LGA/ IdoLGA)
  • Secondary Sampling Unit (SSU): Enumeration Areas (Eight political wards)
  • Elementary units: Households
  • Final Units: Persons

3.3.3.1. Selection of LGA

Oyo state is divided administratively and geographically into 33 Local Government Areas (LGAs), which are subdivided into 361 political wards (Ogundeji 2002). The sample design used the Local Government Areas as strata. Two randomly selected LGAs, an urban (Ibadan North) and a rural (Ido) LGA were selected to create a basis for comparison (see appendix 2 for the map of Ibadan and Ibadan North LGA).

Ibadan North LGA has a population of about 403,513 people (FOS 2001) (see appendix 3 for number of households in Ibadan North LGA per political ward). It is an urban, mostly congested and overcrowded area where most of the inhabitants are traders, artisans and small scale farmers (see appendix 4 for street scenes, houses and waste disposal in Ibadan North LGA). This LGA has one teaching hospital, seven Primary Health Care (PHC) facilities and two general hospitals (see appendix 5 for profile of Ibadan North LGA). The Ido LGA on the other hand is randomly selected out of the nine rural LGA that surround the metropolis of Ibadan. Ido has a population of about 71,529 people (FOS 2001). This LGA has ten PHC centres and most of the inhabitants are small scale farmers (see appendix 6 for profile of Ido LGA). It is important that a rural LGA is selected so that data obtained from the study can be representative of the total population. These two study settings allow for comparison to be made about the actual number of deaths in an urban as compared with a rural setting both in Oyo state.

3.3.4 Sampling frame

The sampling frame should cover 100% of the eligible population in the surveyed country (Fowler 2001; WHO 2004b) this means that every eligible person in the country (or selected region) has a chance of being included in the survey sample. The choice of sampling frame usually depends on using some existing and reasonably up-to-date population lists, or constructing one by undertaking a census of members of households in a defined geographic area (WHO 2004b). Three characteristics of a sampling frame that a researcher should evaluate include comprehensiveness, probability of selection and efficiency (Fowler 2001).

For this study, the electoral rolls were used as a sampling frame since enrolment to vote is compulsory in Nigeria. A household register was also available and used although getting access to this information took considerable time and effort which is the norm in Nigeria when access is needed to any Government documents.

3.3.5. Sample size requirements for sisterhood method

The original work by Graham et al (1989) does not show any method for calculating sample size requirements for surveys employing the sisterhood method to estimate maternal mortality. Graham et al (1989) suggest that “for an estimate of the broad level of maternal mortality, interviews with 3,000-6,000 adults will be required, depending on the expected level of maternal mortality and the number of sisters who can be roughly expected to have reached reproductive age per adult” ( Graham et al ,1989. P: 132).

The WHO however gave a formula to calculate sample size requirement for the sisterhood method:

n = 4P (1-p)/e2where n = number of respondents; P = proportion of respondents to maternal deaths; and e = maximum error that will be tolerated in the proportion, generally 10%. p = average number of adult sisters per respondents divided by estimated lifetime risk of maternal death (WHO 1997c). This formula assumes a simple random sampling scheme. To allow for design effect of using more complex sampling schemes, such as cluster sampling, the derived number of respondents should be rounded up to at least the next 1000 (WHO 1997c). Tables 11 and 12 provide some estimates of sample size requirements for the two approaches compared with household surveys using direct estimation.

Table 11: Number of respondents needed to establish a maternal mortality ratio of 300 per 100,000 live births correct to within 20%

Maternal mortality ratio

Indirect Sisterhood Method

Direct Sisterhood Method

Household Survey

300

4,000*

5,000*

50,000**

*Adult respondents

** Births

Source: WHO (1997c). The sisterhood Method for estimating maternal mortality: Guidance for potential users. WHO: Geneva.

Table 12: Minimum sample size requirements for use of the maternal mortality module (direct sisterhood method)

Maternal Mortality Rate (Maternal deaths per 100,000 women of reproductive age)

Number of DHS adult respondents

20

60,500

60

20,100

100

12,000

140

8,600

180

6,700

220

5,500

260

4,600

Source: WHO (1997c). The sisterhood Method for estimating maternal mortality: Guidance for potential users. WHO: Geneva.

Hanley et al (1996), however analysed nine sisterhood studies (Graham et al 1989; David et al 1991; Chipangwi et al 1992; Oosterhius 1993; Shiferaw and Tessema 1993; Wirawan and Linnan 1994; Walraven et al, 1994; De Groof et al., 1995; Hagen 1995) and used the “yield” (defined as the number of reported sister deaths per thousand respondents) to anticipate the sample-size requirements according to MMR and the desired margin of error.

From the analysis (table 13) it was suggested that “the investigator is required to make an educated guess as to the prevailing MMR in the study and to choose an acceptable margin of error and thereafter choose the needed sample size, from a table that provides sample sizes” (Hanley et al, 1996. P. 223).

Table 13: Approximate sample sizes (numbers of respondents) according to level of maternal mortality ratio and desired margin of error

MMR

Yield

Margin of error

+ 30% +20% +10%

(r> 43) (r>97) (r>385)

250

15

3,000

6,400

25,000

500

30

1,500

3,200

13,000

750

45

1,000

2,100

8,000

Yield = Approximate number of reported deaths per 1,000 respondents.

R= Total number of deaths.

Source: Hanley et al (1996). Confidence interval and sample size calculations for the sisterhood method of estimating maternal mortality. Studies in family Planning. 27, 220-227.

The sample size for this study was therefore determined based on the methodology developed for the Indirect Sisterhood Method (Graham et al 1989). The researcher assumes a maternal mortality ratio of 800 per 100, 000 live births as indicated by UNICEF/WHO in 2000. The number of respondents needed according to the Indirect Sisterhood Method to establish a maternal mortality ratio of ~800 maternal deaths per 100,000 live births per year correct within 20% was ~2,100 household surveys (Hanley et al 1996; WHO 1997c). To account for cluster sampling in the Ibadan North LGA, the researcher included 3,000 respondents in the study. A total of three thousand households were therefore selected from the political wards of Ibadan North and Ido Local Government LGA of Oyo state.

3.3.6. Sampling procedure

The sampling procedure employed in this study followed the principles of multi-stage stratified random sampling, involving four stages. At the first stage, Oyo State had been selected out of the 36 states in Nigeria.

At the second stage, the Ibadan North (Urban LGA) and Ido LGA (Rural LGA) had been randomly selected by balloting as strata. A voters' register/list for the two LGAs was requested from the Oyo state Independent Electoral Commission (OYSIEC) to show the number of people per political wards (See appendix 7 for Voters' list).

At the third stage, household lists in the selected political wards were requested from the National Population Commission. The number of respondent households from each village was calculated using another proportionality factor such that the number of respondent households from each political ward was proportional to the number of households in the political ward. The proportionality factor is stated as follows:

P= n/N*3000

Where,

P= the number of households to be sampled from a political ward

n= the number of households in the political ward

N= the sum of the number of households in the two selected political wards.

The desired total number of households to be selected for the survey, which is 3000.

At the fourth and the last stage, a household numbering was done. This was followed by random selection of households from the list of households.

On average, there were over two thousand adults aged 15-49 in a typical ward (of 2200 households).

Table 14: The total number of households to be selected per political Ward in Ibadan North and Ido Local Government Areas

Local Government Area

Political Ward (abbreviation used in order of coding)

Population (households)

Number of households to be selected

Ibadan North

Gate/Oje/Bere/Aremo (OG)

31150

468

Oniyarin/Inalende (OI)

19950

300

Sango/Ijokodo (SI)

57500

863

Adeoyo/Yemetu (AY)

59500

893

Total

168100

2524

Ido

Akufo (AK)

2510

44

Omi Adio (OA)

19940

350

Elenusonso (EL)

2493

43

Ido (1D)

2200

39

Total

27143

476

In each political ward (table 14), the researcher asked the local village elder or representative to walk the research team around the entire village. All villages were systematically sampled. A sample interval (n) was calculated by dividing the number of households in the cluster or village by the number of interviews to be conducted in the cluster or village. A starting household was determined by random number generation and each nth household was interviewed until the entire cluster or village had been surveyed.

Similarly, in Ibadan North LGA, which is an urban setting, clusters were selected randomly using a map of the area (see appendix 2 for map of the Ibadan North LGA) a house at a particular pre-determined landmark was selected at random from the map and then, according to predetermined protocol, staff selected a house that is a certain number of houses in clockwise direction away from the house. The individual in the chosen age group was interviewed in that house and individuals in the subsequent ten houses adjacent to that house in the specified direction were interviewed also. This constituted systematic random sampling in an urban setting where a list of names is not available since a list of households is available.

3.4. Increasing participation

Non-participation in a survey tends to follow certain patterns and therefore introduces bias or distortion into the data that are collected (WHO 2004b). To increase participation in this study the following were done:

  • The village King or ‘Baale' was paid a courtesy visit by the Researcher, Supervisors and Team Leaders. The aim and objectives of the study were explained, Informed consent sought so as to gain entry to the village and each King was asked to lend their support to the project;
  • A return visit was made to the Kings' palaces to meet with all village chiefs and important dignitaries;
  • A meeting was organised with heads of women's organisations and thereafter their members. Examples of women's organisations visited were: Association of Market Women, association of Women Tailors etcetera;
  • A guided tour of each village was conducted with village guides provided by each village King;
  • Community mobilisation with awareness songs on maternal mortality and the need to prevent it were undertaken in all the villages;
  • Meeting with the Honourable Chairmen of Ibadan North and Ido Local Government Areas;
  • Local Councillors of some political wards within the LGA were also visited; and
  • Several visits were made to the Primary Health Care Coordinators and Chief Matrons of each LGA.

They were all asked to lend support to the survey to promote participation. Media and community groups were also mobilised to increase participation and response rates especially in Nigeria where participation in health surveys is not mandatory by legislation.

3.5. Survey implementation and ensuring quality control

To ensure quality control of this study the following were done:

  • ethical approval;
  • appropriate sample size and methodology were ensured based on the recommended guideline of WHO for conducting surveys;
  • Preparation of the data collecting instrument;
  • translation and back translation of the instrument
  • ensuring cultural adaptations of questions
  • evaluation of translation
  • pilot testing the instrument; and
  • pilot testing all survey procedures

3.5.1. Research Ethics Committee approval

Ethical approval for this study was granted by the Senate Research Ethics Committee, University of Manchester. Ethical approval was also sought and granted by the Oyo state Ministry of Health. A copy of the proposal and ethical approval were submitted to the LGA secretariat for the attention of the Chairmen and Primary Health Coordinators together with cover letters to request permission (see appendix 8 and 9).

3.5.2. Preparation of the data collecting instrument

Since this was a retrospective study, primary data were collected through personal interviews using a structured instrument from individuals at the community level. This means that a questionnaire was developed. This questionnaire (see appendix 10) contained questions on respondents' demographics, demographics of marriage, family and reproductive health and the number of sisters each respondent had. The questionnaire also contained the four basic questions of the Indirect Sisterhood Method (see appendix 11) (Graham et al 1989).

3.5.3. Translation and back translation of the instrument

Surveys in many countries will require that documents be prepared in more than one language. This need may extend beyond instruments and interview schedules to question by question instructions, training manuals, letters of invitation to participants in surveys and materials promoting the project. To ensure accurate reproduction of meanings, it is important that all written materials are translated into the local language of the respondent by at least one individual and then re-translated back to English to check for accuracy and meaning (Bracken and Barona 1991).

Procedure for translation and back translation

  • To ensure validity of this survey instrument, the questionnaire was written in English, and translated into Yoruba, the language of the ethnic group.
  • The translated copy was then given to two individuals who speak and read Yoruba to back translate it into English to assess the questionnaire for accuracy of translation.
  • The retranslated English version and the original English version were compared to see if the meaning of each item was maintained.

The following quality standards were used during the translation and back-translation procedures

  • Translation was aimed at producing a locally understandable instrument.
  • The original intent of the questions were translated with the best possible equivalent term in the local language
  • The instrument was translated by the researcher who has a basic understanding of the key concepts of the study and the study areas
  • The back translation of the instrument was carried out by two separate individuals who are core speakers of the Yoruba language. They both commented on all possible understandings of the terms and suggested alternatives where appropriate.

3.5.4. Ensuring cultural adaptations of questions

In order to ensure cultural adaptation of the instrument, terms and phrases that are culturally appropriate were used. It was ensured also that the translation did not change the meaning and intent of the question; rather efforts were made to make sure that questions retained their original meaning and intent in a culturally appropriate manner. Adequate care was also taken to translate medical terms appropriately into expressions understood by lay people.

3.5.5. Evaluation of translation

Copies of the instrument, the translated and the back translated versions were submitted to the researcher's supervisors. In Nigeria, the two Local Supervisors appointed by the University of Manchester to supervise the Researcher during the fieldwork also evaluated the adequacy of the translated instrument based on their experiences of field surveys and culturally appropriateness of some words and terms. Some corrections were made.

3.5.6. Inclusion criteria

Age range of respondents

The original indirect method based its estimates on the reported mortality of sisters who were or had been married. However, in situations with a pattern of premarital childbearing, this criterion is not optimal and is sometimes replaced by, for example, menarche or reaching the age of 15 years.Restricting the age range of respondents also has an important effect on the MMR estimate. Including respondents over the age of 49 (i.e. beyond reproductive age) extends the time period for the estimate, whereas limiting respondents to those of reproductive age narrows the time period to about 12 years. If we assume that both fertility and maternal mortality have decreased during the time period covered by a study, then it could be argued that the younger respondents would report lower mortality than older respondents (Graham et al 1989). In summary, the age range or respondents for this study was restricted to 15-49 years which is the reproductive age.

Gender of the Respondents

All adults, either male or female were included in the study. Although it may be expected that women give more accurate reports on their sisters than do men, particularly with regard to child bearing matters, this is likely to vary between societies. In Nigeria, men are well informed and involved in their sisters' lives. Involving men also helped to prevent loss of information on deaths of sisters from one female families.

3.5.7. Pilot testing the instrument

It was necessary to pilot test the translated instrument on a few people with a broad range of backgrounds including low and middle socio-economic status to ensure the validity and reliability of the instrument.

3.6. Pilot Study

This pilot study was conducted between 11th April and 22nd April 2005. Four hundred and twenty houses were visited and interviews of 420 respondents were conducted in a randomly pre-selected Local Government Area of Oyo state not included in the main study.

Objectives of pilot study

The objectives of the pilot study were to:

  • test adequacy of research instruments
  • improve research techniques;
  • determine an appropriate workload;
  • determine the time required for interviews;
  • assess the feasibility of a (full-scale) study/survey;
  • assess the willingness of people to take part in a study; and
  • report on other substantive findings arising in the pilot.

3.6.1. Study Setting

The setting for the pilot study was the Ibadan North east Local Government Area of Oyo state. Ibadan, the largest indigenous city in tropical Africa, is the capital of Oyo state, which is one of the 36 states of the Federal Republic of Nigeria (excluding the Federal Capital territory, Abuja). It is centrally situated in the South-western sector of the country, and is 145km north-east of Lagos and 372km south-west of Abuja. Its central location gives it transport and economic advantage, as is shown by the five primary roads and the expressway from Lagos which converges on it radially. In addition, the railway from Lagos to Northern Nigeria passes through the city. It is thus in a commanding economic position, which to a large extent explains its rapid growth. At present, the exact spatial coverage of the built-up areas is unknown. For administrative purposes, the city is made up of 11 Local Government Areas (five in the inner urban area, and six in the outer rural area). It has an estimated population for the year 2000 of well over 3,000,000.

3.6.2. Research Supervision during pilot study

Overall monitoring of the research was undertaken by the researcher. This provided opportunity for on-the-spot discussion with supervisors and interviewers of problems encountered in the field, clarification of questions with regard to the questionnaire, assessing questionnaires for completeness and distribution of supplies. On a daily basis, completed questionnaires were collected and edited by the researcher. From this editing questions that were not clearly understood were identified and further instruction was provided as necessary.

The research Supervisors were responsible for allocating daily work to Data Collectors, checking and registering forms, collecting and correcting missing and conflicting data and ensuring that Data Collectors developed adequate rapport with respondents. In addition, each Supervisor conducted a 5.0% validity check of houses/respondents and their questionnaires in the Local Government Area. These houses were randomly selected by the Researcher without the knowledge of Data Collectors. The purpose of the validity check was to monitor the reliability and consistency of the Data collectors.

3.6.2.1. Repeatability check by reinterviewing

Immediately after the pilot survey, 5% of questionnaires were randomly selected for reinterviewing. The main goals were to assess the quality of the interview data and to make sure the interview was conducted properly. Of the 420 respondents during the pilot survey, 21 questionnaires were selected using systematic random sampling through the use of a table containing all the questionnaire numbers. Selected respondents and their locations were randomly allocated to supervisors. A repeat of all items contained in the questionnaire was done. Responses to questions on the number of sisters, marital status, name of sisters and information about dead sisters were quite consistent. There was however, inconsistency for questions on causes and prevention of maternal mortality in the community as respondents provided more information on these questions. Supervisors also reported a shorter time for completing the questionnaire which might be due to the fact that respondents have become more familiar with the questions. Research supervisors reported that some respondents resented being reinterviewed and felt that they were being “checked up on” and one respondent refused to be reinterviewed. The data were fully coded using standard procedures used in the pilot survey and the result of data checks are in appendix 12.

3.6.3. Pilot study procedures to improve the internal validity of a questionnaire

  • The questionnaire was administered to pilot subjects in exactly the same way as it was to be administered in the main study.

In each selected house, the identity of the Data Collector, and the reason for the visit were made clear. Verbal informed consent was sought and the respondent was informed that he/she had the freedom not to participate in the study or not to continue answering the question at any stage of the interview. Data Collectors asked subjects for feedback to identify ambiguities and difficult questions. As the Data Collectors interviewed respondents, respondents were asked if each question was clear. The Respondents were asked in particular if the framing and wording of the translation into the Yoruba language were adequate and if not they were asked for suggestions in order to make each question culturally acceptable. No correction was raised as regards the wording, structure and translation of the questionnaire. Only one respondent raised the issue that Data Collectors should show more sympathy as the subject is referring to someone very close, a “dead sister”. This was noted, reported and further discussed in the classroom. Data Collectors were also asked to take their time in filling the questionnaire and not “rush through”. The aim was to collect all information and not on the overall number of the questionnaires that Data Collectors could complete per day. Indirect questioning, sometimes in different ways was also used to elicit appropriate answers.

  • the time taken to complete the questionnaire was recorded

The time taken to complete each questionnaire was an average of 60 minutes, including the time for Interviewer's introduction. Considering the sensitivity of the research and the fact that there were 30 items in the questionnaire it was decided that the time was reasonable.

  • Efforts were made to check that all questions were answered

Modalities were put in place to establish that replies could be interpreted in terms of the information that is required, as well as to check that all questions were answered.

  • Determining appropriate workload

More was learned about the environment and feasibility of the survey because it gets very hot and sunny unusually early and hampered progress and this could be expected to also affect the main study. It was also realised that due to these adverse weather conditions and the need for data collectors to be more sensitive, which makes the completion of the questionnaire to take an average of 60 minutes, each data collector can conveniently complete two-three questionnaires a day.

During the pilot testing of the instrument, Interviewers asked pilot respondents:

  • If they understand all the words?
  • If they know what is being asked, this is to check how clear the intent of the question is.
  • If they have any questions about it
  • How the question could be made clearer
  • If there is any questions that make the respondent feel uncomfortable. If there are such questions how best the respondent feels such questions can be asked.

3.6.4. Outcome of the pilot study

Data Analysis

All questionnaires were reviewed for completeness and correctness of recording on a daily basis. Data coding was done and completed data were entered using excel. Data were analysed using the statistical package SPSS.

Results

A total of 420 Respondents were invited to participate, all did. This yields a response rate of 100%. Two hundred and seventy two respondents (64.8%) were female and 148 (35.2%) were male. The age range of respondents was from 15-60 years with a mean age of 33.01 and standard deviation 8.663. Most of the respondents (308, 73.4%) were married, 5 (1.2%) were widowed, 106 (25.2%) were single and 1 (0.2%) was divorced.

a) Maternal Mortality

Respondents provided information about 421 sisters. Of the 421 sisters 108 (25.7%) were dead. Eighty-seven (80.5%) of these deaths were maternal deaths with 17 (19.5%) deaths during pregnancy, 40 (46.0%) deaths in childbirth and 30 (34.5) deaths during puerperium.

b) Place and Time of death

Respondents reported that 13 (14.6%) of their sisters died at home and of these home deaths two died during pregnancy, three during labour, none died in childbirth but eight died after delivery. The majority of these deaths 35 (39.3%) occurred at private clinics with a massive 16 deaths after delivery and 11 during labour. Twelve (13.5%) of the deaths occurred on the way to Hospital after five of the sisters had already delivered. A substantial number of deaths 24 (27.0%) also occurred in the Hospital, 14 of these patients were already delivered. Three (3.4%) of the deaths occurred in a church/Mission House.

c) Time interval between delivery/termination of pregnancy and death

Fifty (57.5%) out of these 87 deaths occurred within 24 hours of delivery or termination of pregnancy, 17 (19.5%) within 1-3 days, 9 (10.3%) within 4-7 days and 11 (11.5%) between 8-42 days of delivery/termination of pregnancy.

d) Admission-death interval

Although 65 (60.2%) of the sisters were reported to have been admitted before death, 21 (19.4%) of those dead were never admitted to any health facility. Information about one sister was not provided. Of those admitted 35 died within 24hours of admission, 17 died within 3 days, 8 died within 4-7 days, and 7 died within 42 days of admission.

e) Cause of death

The cause of deaths of the sisters as reported by the Respondents were as follows: 30 (34.1%) were due to haemorrhage, prolonged/obstructed labour accounted for 14 (17%), infection accounted for 13 (14.8%), 7 (8.0%) deaths were due to anaemia not due to haemorrhage, 4 (4.6%) to eclampsia/preeclampsia, abortion accounted for 4 (4.6)% of the deaths, 3 (3.7%) died during or after caesarean delivery, 2 (2.3%) to uterine rupture, and 1 (1.1%) death was due to Jaundice which may be as a result of Sickle cell disorder. Information about the cause of death or the signs and symptoms manifested by the deceased sisters were not provided for 12 deaths. The 21 deaths not due to maternal causes were reported to have happened during infancy or at old age.

3.6.5. Lessons Learnt from Pilot Study

Willingness to participate in the study

The pilot was supported by the Director of Primary Health Care and the Measurement and Evaluation unit of the LGA. Approval for the study was sought from the Oyo state Ministry of Health Ethical Committee and the Chairman of the Ibadan North East LGA. The study was also approved by the University of Manchester Senate Research Ethics Committee. The cooperation of the public was excellent and participants were willing to participate in the study. The response rate for the study was 100%.

Uncovering of Local Political problem and other issues

This pilot study also uncovered local political problems and other issues that may be encountered during the main study. Before the commencement of the pilot study, the Researcher had visited the Monitoring and Evaluation Unit of the LGA, going through the records, there was no incidence of maternal mortality in the political wards selected for the pilot. The Researcher also visited Government Health Centres in the Locality and just like the Headquarters there was no incidence of maternal mortality in the area. Astonishingly, on the first day of the pilot, five cases of maternal deaths were identified in two streets of the political ward. One of the deaths had just occurred a few days before the data collection and the respondent stated that the death occurred in the Health Centre. The Researcher made a return visit to the Health Centre since there was no record of any death in the Health Centre. It was then discovered that the deceased, had actually been admitted in labour to the centre. When her labour was unduly prolonged, the Midwife referred her to Adeoyo Maternity Hospital (a secondary Health care facility). Whilst the family were transporting this woman to the hospital they met a family friend who encouraged them to visit an Herbalist who had helped the friend when she had similar experience in labour. The husband then took his wife there. By the time the family realised that the Herbalist could not help, this woman was already in a critical condition. Knowing that the Hospital was a long distance away, the family decided to bring the woman back to the Health Centre. The Midwife refused to readmit her because, according to the Midwife, this woman had been referred. This woman eventually died in the car. The Health Centre did not record this death because she did not die in the Health Centre, but outside. This case was reported to the Director of Primary Health Care to prevent this happening in the future.

3.6.6. Implications of the pilot on main study

The pilot study had the following implications on the main study:

  • Probably due to the extensive time used for the class room training, role play, review of questionnaire and practice on the field as well as the constant supervision of the data collectors, no mistakes were made in the completion of the questionnaire
  • Respondents were asked if the questions and translations were well worded and to state any question that appeared confusing or ambiguous. Throughout the pilot study no observations or correction was made as regards the wording or translation of the questionnaire. As a result, it was not necessary to make any changes to the layout, wording or translation of the questionnaire;
  • The study was also well received and all respondents were willing to participate in the study (100% response rate)
  • More was learned about the environment and feasibility of the survey because it gets very hot and sunny unusually early and hampered progress, and that could be expected to affect the main study;
  • Some respondents commented that the study was on a sensitive issue (the death of a sister) and that data collectors should be more tolerant and patient and not rush through the interview;
  • During practice debriefing it was emphasised to data collectors that tolerance and patience were key points to ensure the success of the survey; and
  • In spite of these adverse weather conditions and the need for data collectors to be more sensitive, it was discovered that it takes an average of 60 minutes to complete a questionnaire and that each data collector can conveniently complete two-three questionnaires a day.

3.7. Data Collection

3.7.1. Staffing/Recruitment of Data collectors, Team Leaders and Research Supervisors

Ten student Midwives of the Oyo State School of Midwifery and the University College Hospital School of Midwifery were recruited as data collectors. The intention was to recruit 20 student midwives but at that time of the school's academic year, the new set of students were just resuming and the previous set had just finished their qualifying examination and some of them had secured permanent jobs. Therefore, 10 newly qualified midwives were recruited as data collectors. It is thought that recruiting midwives would be advantageous to this study as they are the health professional primarily concerned with the care of women during pregnancy and childbirth so as to have “real life” experience of the incidence of maternal mortality. Since 20 Data Collectors were needed, 10 experienced data collectors from the Faculty of Public Health, College of Medicine, University of Ibadan were also recruited. Two Team Leaders and two Supervisors who had previously been involved in this type of research were also recruited to serve as research team leaders and field supervisors. The two Supervisors and one of the Team Leaders recruited were experienced field supervisors and Team Leader working at the Faculty of Public Health of the College of Medicine, University of Ibadan.

3.7.2. Training

A four-day training was planned consisting of two days classroom teaching, one day practice and role play to review the questionnaire in the class and one day practice on the field. During training it was realised that a day would not be enough for practice in the classroom to ensure that a reasonable level of competence had been attained by data collectors hence two days were used for practice on the field (see appendix 13 for programme of training). The training therefore was for five-days consisting of: two days classroom teaching on specific issues, such as general overview of maternal mortality and the sensitivity of obtaining data on the subject at the community level; two days practice and role play on the review of the questionnaire in the classroom; and a one day practice in the field, to ensure “hands on” experience on the use of the questionnaire. At the end of this day the field practice was followed up with classroom debriefing of the students' experiences and difficulties. The combination of classroom teaching and practice interviews in the field, followed by debriefing and general discussion, was crucial to the motivation of the Data Collectors and the success of the study (Kwast et al 1986).

The purpose of the training was to

  • orient Data Collectors and Supervisors to the problems of maternal mortality;
  • explain the rationale of the study and the study protocol;
  • specify the survey objectives;
  • discuss the importance of the data to be collected;
  • explain the importance of the Data Collectors and Supervisors;
  • explain and promote appropriate interviewing techniques;
  • ensure a full understanding of every item of the questionnaire;
  • ensure uniform application of the survey instrument;
  • improve the overall quality of the data; and
  • motivate Data Collectors

A training manual (see appendix 14 for the training manual) was prepared for the field staff, containing the questionnaire together with the instruction on the method of completion. It comprised four chapters:

  • Overview of maternal mortality;
  • Aims and coverage of the survey;
  • Data collectors, Team Leaders and Supervisors Tasks; Ethics and obtaining informed consent; and
  • Instruction on the completion of the questionnaire

The manual was prepared in English and its aim was to serve as a reference guide for Data Collectors and Supervisors in the field. In-depth instruction was given on the sequence of collecting data using the sisterhood approach and special attention was paid to the word “sister” during teaching as this word could mean nieces, cousins and aunts in the Nigerian context, but with the Sisterhood Method it only includes sisters born of the same mother.

3.7.3. Research Supervision

Overall monitoring of the research was done by the researcher. This provided an opportunity for on-the-spot discussion with supervisors and interviewers of problems encountered in the field, clarification of questions with regard to the questionnaire, assessing the questionnaire for completeness and distribution of supplies. The researcher also had a University Supervisor locally in the Faculty of Public Health in the University of Ibadan as well as an experienced field supervisor based in a Non Governmental Organisation in the community. The researcher's primary supervisor from the University of Manchester also visited during data collection and was constantly available via e-mail.

On a daily basis, completed questionnaires were collected and edited by the researcher. From this editing questions that were not clearly understood were identified and further instruction was provided as necessary. The research field Supervisors were responsible for allocating daily work to Data Collectors, checking and registering forms, collecting and correcting missing and conflicting data and ensuring that Data Collectors developed adequate rapport with respondents. In addition, each Supervisor conducted a 5.0% validity check of houses/respondents and their questionnaires in the LGA. These houses were randomly selected by the investigator without the knowledge of the Data Collectors. The purpose of the validity check was to monitor the reliability and consistency of the Data collectors.

3.7.4. Team leader/supervisor's tasks

  • Ensured all survey tools were available;
  • Ensured existing cultural and traditional protocols were followed prior to the survey;
  • Solved irregularities in household listing and assisted with identification of household;
  • Checked instruments for completeness and errors on a daily basis;
  • After each survey site visit, the team leader was responsible for recording the number of instruments completed and submit to Supervisors who in turn checked each questionnaire for completeness and submitted all questionnaire to the survey office; and
  • Be accessible to interviewers at survey base to solve pending problems.

3.7.5. Interviewers' tasks

  • Obtained verbal informed consent;
  • Conducted interviews;
  • Made a primary check of the instrument before handing it over to the team leader; and
  • Reported to the team leader any difficulties encountered during interview.

3.7.6. Conducting interviews in the field

3.7.6.1. Gaining access to study setting

A natural setting could be “closed” or “invisible” (Sim and Wright 2000) if the information is closed to public scrutiny, e.g. in Board meetings of organizations and open or visible if information is available to the general public e.g. information centre. Access can only be gained to a closed setting by permission from the authorities who are known as “gatekeepers” (Hammersley and Atkinson 1995). To gain access to a community, a researcher has to go through a respectable member of the community (Fetterman 1998). In Oyo state, approval of the study was obtained from the State Ministry of Health. Written permission (see appendix 8 and 9 for letters of permission from Ibadan North and Ido LGAs) was also sought from the chairmen of the Local Government Areas. The local chiefs of the study communities were consulted and their consent gained, as well as the administrators of health facilities. In each selected house, the identities of the researcher and interviewers were made clear, and explanation of the visit was given.

3.7.6.2. Approaching the public

Interviewers were asked to clearly communicate the objectives of the survey to the respondent. S/he established a good rapport by introducing her/himself and the survey.

As a result of the low literacy level, the questionnaire was not administered to the study respondents but interviewers read out questions to the respondents and documented responses from the respondents. At the bottom of the questionnaire, the data collector asked if there was anything, which the respondent wanted to ask her/him. This was very important and it ensured that communication during the interview was not one sided or interviewer dominated.

Collection of data was done by asking the participants valuable questions and by seeking their cooperation. Each interviewer had a name tag which was worn at all times. They were trained to be tactful and not to force respondents to participate in the study, but be prepared to come at another time. They were asked to always stress the confidentiality of the information and make sure that respondents felt comfortable about the survey. They also informed respondents that they could refuse to participate in the survey since participation was voluntary.

3.7.6.3. Respect for Autonomy: Gaining Informed Consent

Informed consent is the process by which the risks, benefits, and expectations of an intervention or research project are disclosed to a patient or participant in order for the participant to make an informed decision concerning acceptance or participation (Coughlin and Beauchamp 1996). Ethical principles of Western medicine on informed consent requireall adults to be primary decision makers about their own participation.The researcher must obtain informed consent from each research participant and written consent is signed and kept by the participants with the understanding that confidentiality and anonymity will be maintained (Burns and Grove 1997).

The utility of written consent is called into question when participants cannot read (Upvall and Hashwani 2001). Asking participants to also sign any document (including a consent letter) when they cannot read or fully comprehend its contents, can be threatening to participants who only sign or use thumb prints for marriage documents or other significant events in their lives (Upvall and Hashwani 2001). More importantly, participants may see documents as tax forms since they cannot read the content thereby declining to participate in the study. This means that the relevance and applicability of Western principlesof autonomy and individualism need to be contextualised to thesesettings (DeCosta, D'Souza, Krishnan, Chhabra, Shihaam and Goswami 2004).

The adaptation of research ethics to the cultural values of the community has often been debated by several authors (Ajayi 1980; Christakis 1992; Barry and Molyneux 1992; Upvall and Hashwani 2001; DeCosta et al 2004; Masiye 2006). A community based approach to enrolment has been suggested by some authors whereinthe decision regarding participation is taken by the village/communityleader, since decisions are taken at the community rather than the individuall level in some societies (Ajayi 1980; Upvall and Hashwani 2001; DeCosta et al 2004). In this study, the individual rights of each study participant were protected by seeking approval from the research ethics committees overseeing the research from the University of Manchester and the state Ministry of Health.

Verbal informed consent was sought from the study participants. This was done by going through the community head of each study village, seeking the consent of the village heads who then met together with the researcher with the local chiefs initially and the entire village. This was done because access to individuals cannot be gained, and consent cannot be sought in African communities without going through the community leader. Also, field investigations in developing societies such as Nigeria take place within a particular context of regional values and practices, local concepts of disease (and health), and power hierarchies in the family/community that constitute local cultural systems.

In spite of the fact that the community leader was used to seek the consent of the community and individual participant, the researcher ensured that every participant was free from coercion. Participants were told that they were not obliged to participate in the study and that they were free to withdraw at any stage of the study.

3.7.6.4. Making contact

Contact attempts were made by actually knocking at the door of the household. Where household members were not available at the first visit, Interviewers had to make at least three different visits to a household or eligible respondent when trying to obtain an interview. Scheduling call backs at different times was important and helped in reducing the rate of non response.

3.7.6.5. Functioning of Teams

The cooperation within the teams remained good throughout the survey and Data Collectors and Supervisors related well to each other. At the end of each month, on pay day, Data Collectors and Supervisors were asked whether they wished to continue with the survey as it was essential that changes and replacement, if any, should take place at that time. There was no personnel attrition.

Initially the members of the Teams were transported in a private car. However, the car had to make several trips a day to get all the members of the team. Also, due to very poor road network, to the site especially in the rural area, there was the need to change to public transport. Public transportation worked well for the majority of the villages except in the following:

  • Elenusonso Village: Transport was difficult to get and a lot of time was wasted. The survey time had to be extended, transport allowance was increased and the data collecting team had to use a taxi. When there had been rain the previous night, access to this village was not possible the next day because there is a river on the way to the village with a wooden “Make Shift Bridge”. With any amount of rain the river overflows and vehicles cannot get in or out of the village. On one occasion when the river overflowed the team had to get out of the taxi, tried to walk over the bridge and it took them about an hour to walk to the village. This was very risky as the bridge was made with tiny planks which could break at anytime. Anytime there was rain, data collection for that day in this village had to be cancelled and Data Collectors moved to the next village.
  • Akufo Village: Transport was a problem. The road was filled with giant potholes and a truck had to be used to get access to the village. Also villagers are farmers and they leave home as early as five o'clock in the morning but could only be met at home on market days. Data collection had to be changed to market days which were twice in a week. Here again, the survey time was extended and extra transport allowance had to be provided.
  • In Ibadan North LGA, sometimes Data Collectors worked in the evening as most adults work in offices and shops outside the home and only children could be found during the day.

At the start of the data collection in a new area, the team familiarised itself with the layout of houses in the political ward with the assistance of a field guide provided by the village head or the LGA secretariat.

3.7.6.6. Editing

Immediately after the completion of the questionnaire the interviewer reviewed the questionnaire to check for completeness and that no question was omitted.

3.7.6.7. Practicalities of data collection

Practicalities of data collection have been discussed to some extent above but it is important to further highlight those practicalities and further discuss each point since these practicalities have implications for the quality of the study:

  • There is a tendency for design bias. This is because data collection was undertaken in one urban political ward which required the use of complex sampling techniques such as cluster sampling. In order to overcome design bias, larger sample sizes were used in the urban LGA (>2000 respondents).
  • It was important to carefully pre-test the questionnaires and to pay special attention to translation into local languages, particularly of the term "sister" which is understood differently in different cultures. Therefore, the qualification “born to the same mother” was stressed during the training of data collectors and data collectors were asked to pay special attention to the use of the term during data collection periods. This qualification “born to the same mother” helped to overcome some of the ambiguities in the term “sister” which in Nigeria may be used differently for a wide variety of relations including aunties, cousins and nieces. Also in Nigeria owing to polygamous marriages, individuals may have the same mother but different fathers and vice versa the word “sister” is used as a general term of endearment. For this study, the term only included sisters born to the same mother.
  • It was also particularly important to ensure that female respondents did not include themselves among the sisters reported as this would inflate the denominator and thus deflate the true maternal mortality ratio.
  • Good training and careful supervision of data collectors during data collection was given high priority. As stated earlier overall supervision was done by the researcher and each research team was initially visited daily. Although this was expensive and difficult, it appeared as the only available option to ensure constant monitoring at the beginning of the study. The researcher later visited thrice weekly and the use of mobile phones was very helpful for use between field supervisors and the main researcher but this had implication on the cost of the fieldwork.
  • Before data collection, the researcher acknowledged that the indirect sisterhood method might not be able to accurately capture abortion-related deaths, especially as a result of the illegality of abortion in Nigeria and the adoption of Sharia law in some parts of Nigeria. But during data collection respondents freely gave information about their sisters who died of induced abortion.

3.8. Validation of cause of death

Acquiring information on the causes of maternal mortality is important in determining the public health importance and priorities of specific maternal health problems; in designing appropriate interventions to reduce maternal mortality; and in evaluating the impact of health interventions on maternal mortality (WHO 1994). Ascertaining the cause of death in women of reproductive age is critical because death during pregnancy, childbirth and puerperium constitute a large proportion of avoidable or preventable death in developing countries (Airede 2004). Accurate information on the nature and prevalence of causes of death is essential to guide policy-makers in establishing priorities for Safe Motherhood programmes (WHO 1994 p: 1).

Ideally, when a death occurs, a death registration form is filled by a physician who lists the conditions or diseases leading to the death and it is conventional to assign a single underlying cause of death on the death certificate (WHO 1994). Data on causes of death are usually obtained through routine vital registration of deaths from the death registry, in developed countries death registration is universal but in situations where access to medical care is limited and where a significant proportion of the deaths occur at home in the absence of trained medical personnel as is the case in developing or resource poor countries ascertaining the causes of death may be difficult but is important (WHO 1994). Since the lack of such crucial information presents various problems in prioritising and evaluating health programmes. This has led to the use of retrospective interviews to ascertain cause of death in developing countries refer to as verbal autopsy (Chandramohan, Maude, Rodrigues and Hayes 1994; Chandramohan, Rodriques, Maude and Hayes 1998a; Chandramohan, Maude, Rodriques and Hayes 1998b).

Approaches to validating the cause of maternal death in this study involve the following: development of an interview instrument for validation; incorporating local disease concepts; and the use of an external validator (Kalter 1992).

  • Development of an interview instrument for validation

Diagnosing an illness from an interview is based on respondents' report of the presence of signs and symptoms which accurately predict the diagnoses (sensitivity) of a condition while excluding the other diagnoses which are truly absent (specificity) (Gray, Smith and Brass 1990; Zimicki 1990; Kalter 1992; Chandramohan, Setel and Quigley 2001). Effectiveness of this method depends on the actual presence of a distinguishing sign, the initial recognition and subsequent recall of these signs by the respondents (Snow and Marsh 1992).

The first step in the development of a structured interview to validate the cause of death in this study consists of: using a structured instrument (Kalter 1992); defining signs known to be associated with each cause of maternal death which may or may not occur with other diseases; developing a characteristic spectrum of signs which always occurs with the disease; and including a sign which always occur with a disease. This has the potential to identify all cases of the disease (100% sensitivity) and exclude all other possible diagnoses (100% specificity); as well as including the timing and duration of signs.

  • Incorporating local disease concepts

The questionnaire was translated into Yoruba the dialect of the study setting and local disease terminologies were included in the questionnaire. For example the inclusion of the Yoruba concept “Oyun baje mo lara” have greater specificity for spontaneous abortion and helped to prevent confusion with unsafe abortion during data collection. This approach, although made the questionnaire lengthy offers the advantage of simultaneously adjusting for biological factors and for recognition and recall within the particular cultural context

  • The use of an external validator

Validation of a mortality interview requires comparison of interview-based diagnoses with some external standard. Ideally, the standard is a direct comparison with medical (Kalter, Gray, Black and Gultiano 1990; Chandramohan 1998a; Quigley, Chandramohan and Rodiques 1999); or post mortem diagnoses (Kalter et al 1990; Chandramohan et al 1998b; Chandramohan et al 2001). Direct comparison enables the determination of the sensitivity and specificity of individual signs and combinations of signs reported by a respondent (Kalter 1992).

Validation of the cause of maternal death in this study was undertaken by a validator who was independent of the study. This process provided an indirect method of ascertainingcause of death from information about symptoms and signs obtainedfrom bereaved relatives. Diagnoses were assigned on the basis of a review of questionnaire, a fulfilment of disease specific algorithms (combinations of signs and symptoms) and the presence or absence, timing and duration of signs and symptoms.

3.9. Analysis

The data are descriptive, as this was the nature of the study. All the data were coded and entered into the computer using excel worksheet (Excel 2003). Data was double entered to detect errors in data entry. The Researcher's second supervisor (Dr. Malcolm Campbell) helped to read the Excel data into SPSS. Analysis of the data was done using the Statistics Package for the Social Sciences (SPSS 14) software. Most of the analysis is descriptive as stated earlier, reporting counts, percentages and summary statistics. Some of the frequencies were cross-tabulated by strata and clusters. Some statistical tests were run later, Pearson's chi square test or Fishers Exact Test and T- test. These statistical tests are important in order to detect any difference in proportion based on stratification and to ensure such difference is not due to chance where it exists. For simplicity also, the tests assume simple random sampling and do not take stratification or clustering into account. To estimate the Total Fertility Rate and the Maternal Mortality Rate two Excel spreadsheets were created by the Researcher's second supervisor. Statistical advice was obtained from the Researcher's second supervisor and from the statistician of the first author (Graham W.) who proposed the indirect sisterhood method in the University of Aberdeen.

Summary

This chapter has described the methodology of this study. Emphasis has been placed on sampling method, sampling procedure and the pilot testing of the instrument as well as the pilot testing of the entire survey procedure. This chapter also discussed the data collection and the training of the research team. How access was gained and issues of autonomy and gaining informed consent were also discussed. The findings of the study are discussed in the next chapter.