correspondence Dr Allan G. Hill, Dept. of Population and International Health, Harvard School of Public Health, 665 Huntington Avenue, Boston MA 02115, USA. E-mail: AHill@hsph.harvard.edu
Summary Using data from a longitudinal study conducted in 40 villages by the UK MRC in the North Bank Division of The Gambia beginning in late 1981, we examined infant and child mortality over a 15-year period for a population of about 17 000 people. Comparisons are drawn between villages with and without PHC. The extra facilities in the PHC villages include: a paid Community Health Nurse for about every 5 villages, a Village Health Worker and a trained Traditional Birth Attendant. Maternal and child health services with a vaccination programme are accessible to residents in both PHC and non-PHC villages. The data indicate that there has been a marked improvement in infant and under-five mortality in both sets of villages. Following the establishment of the PHC system in 1983, infant mortality dropped from 134/1000 in 1982–83 to 69/1000 in 1992–94 in the PHC villages and from 155/1000 to 91/1000 in the non-PHC villages over the same period. Between 1982 and 83 and 1992–94, the death rates for children aged 1–4 fell from 42/1000 to 28/1000 in the PHC villages and from 45/1000 to 38/1000 in the non-PHC villages. Since 1994, when supervision of the PHC system has weakened, infant mortality rates in the PHC villages have risen to 89/1000 in 1994–96. The rates in the non-PHC villages fell to 78/1000 for this period. The under-five mortality rates in both sets of villages have converged to 34/1000 for 1994–96. When the PHC programme was well supported in the 1980s, we saw significantly lower mortality rates for the 1–4-year-olds. These differences disappeared when support for PHC was reduced after 1994. The differential effects on infant mortality are less clear cut.
In the 1980s considerable efforts were made to establish low-cost community-based health programmes in poor countries following the Alma Ata declaration of 1978 ( WHO 1978). Some early reports were encouraging ( Berggren et al. 1981 ) but a recent review of the impact of different health interventions suggested that evidence for the impact of such interventions on the mortality rates of young children is weak ( Ewbank & Gribble 1993). A few of the more carefully studied interventions have suggested that broadly based approaches such as Primary Health Care (PHC) have been relatively ineffective ( Greenwood et al. 1990a ,b; De Francisco et al. 1994 ; Magnani et al. 1996 ; Jaffar et al. 1997 ), whereas disease-specific interventions, especially vaccination programmes, appear to have had a clearer impact on mortality ( World Bank 1994; Desgrées du Loûet al. 1996 ). The evidence so far on the effectiveness of PHC is rather slim although case studies of implementations are more common ( Cham et al. 1987 ; Kasonde & Martin 1994; Chabot et al. 1995 ). There is thus an urgent need to have a clearer understanding of the impact of comprehensive PHC on mortality and morbidity to avoid erroneous policy decisions in the future.
Assessment of the independent impact of any large-scale intervention in the field rather than laboratory conditions is fraught with difficulties. When dealing with an intervention such as Primary Health Care, which by its very nature is meant to deal simultaneously with several health problems, the identification of the particular contribution of PHC to specific conditions is complex. In such cases, recourse to overall mortality as an outcome is preferable to the use of cause-specific mortality and morbidity ( Mosley 1988; Mosley & Becker 1991; Chen et al. 1993 ; Becker & Black 1996). When so much is at stake in terms of the policies to be adopted for the provision of health services in poor countries, it is important to fully exploit any opportunity to clarify the contribution of PHC to past health improvements and its role in the provision of health services in the future. In this task, data from longitudinal studies are invaluable, since cross-sectional DHS-type data have been shown to be an imperfect lens through which to search for causes and effects in the health sector ( Hill 1992).
We used data collected over a 15-year period in 40 villages in the North Bank Division of central Gambia, and employed a mixture of demographic and epidemiological arguments to interpret them. The surveillance system was established just before 1983 and evaluation of the PHC programme was one of the main initial reasons for setting up the system.
PHC in The Gambia
>Around the world, the content and procedures for implementation of PHC have varied widely ( Bobadilla et al. 1994 ). Mosley (1985) drew attention to the five components agreed at Alma Ata, pointing out that in many instances, only the last two had been enacted:
Active community participation: control over the content and implementation of PHC;
Social relevance: Western medicine complements local systems;
Other sectors involved: education, water and sanitation, agriculture, etc.
Health service and health promotion operated by paramedical personnel;
Identification of larger ‘key’ villages to receive Community Health Nurses (CHNs) trained by the Medical and Health Department and paid from central funds. CHNs form the link between village-level primary health care services and referral health services available at dispensaries and health centres. Each of the CHNs is responsible for the supplies, supervision and the continuing education of the village health workers (VHWs) and trained birth attendants (TBAs) in about 5 PHC villages.
Encouragement of villages with over 400 residents, ‘PHC villages’, through their Village Development Committees to identify suitable persons for training as VHWs and traditional birth attendants for training as midwives, to generate resources within the community to contribute to the support of the village health system including VHWs and TBAs, equipment, and replacement of base-line drugs, e.g. chloroquine, aspirin, Penicillin V and oral rehydration salts (ORS).
Development of an Expanded Programme of Immunization and maternal and child health services originally free of charge and since 1988 with a flat fee system. A regular rota was established for a mobile team from the District Headquarters to visit the key villages. At these clinics, the local CHNs, VHWs and TBAs work together with the district health staff, and the target population includes residents from the surrounding PHC and the smaller non-PHC villages nearby.
Village-level health services developed rapidly in the early 1980s, as the 1984–85 PHC review revealed. By the time of this assessment, there were over 200 villages with village health services, village health workers and trained birth attendants ( Chambers 1986). Already, differentials in clinic attendance for antenatal or postnatal care were showing up. In the PHC villages, over half of the deliveries were by trained personnel compared with 15% in non-PHC villages. In the PHC villages, knowledge of vaccination was higher, the value of solid food for the diet of the young child more widely appreciated and the use of oral rehydration therapy better known than in the non-PHC villages ( Chambers 1986).
In 1997, there were 402 PHC villages in the country, 460 Trained Birth Attendants, 52 assistant TBAs and 526 Village Health Workers ( Gambia/UNICEF 1997). Between 50% and 60% of the population have access to PHC services. Trained health workers including TBAs conducted half of all deliveries in 1997. In 1989, a Village Health Worker ( Gambia/UNICEF 1997) had attended 40% of all outpatients at government health centres.
PHC has thus been taken seriously by The Gambia and the fraction of the health budget for PHC and basic health services has been defended since its inception. In 1978, the hospital services (mostly the Royal Victoria Hospital and Bansang Hospital) consumed more than 70% of health sector resources ( Gambia/UNICEF 1997). The balance was in favour of PHC and basic health services in the 1980s but some gains were lost in the early 1990s. Over the 1992–96 period, recurrent allocations for the two hospitals were a little less than the total spent on PHC and basic health services, indicating some reduction in the priority attached to PHC in the 1990s ( Gambia/UNICEF 1997). In addition, the development of the Farafenni hospital and plans for an additional hospital in Bwiam are further threats to the proportion of the health budget spent on village-level health services.
The principal providers of the funds for PHC are the international donors (72% is the forecast: Gambia/UNICEF 1997), leaving the government to pay the rest from its own resources and for most of the tertiary services. Overall, external donor support covered over 60% of total health expenditures during 1991–4, falling slightly to 55% in 1994–6 ( World Bank 1997) ( Table 3). Indeed, over one-fifth of Gambia's GNP has stemmed from foreign aid ( World Bank 1997) ( Table 3). After the 1994 coup, overseas donor support halved in 1995 but was still estimated to be about US$ 6.50 per head, the bulk destined for support of primary health care (28% of total assistance) and for the immunization programme (23%) ( Department of State for Health & Social Welfare Gambia 1998). Despite these vicissitudes, it is clear that PHC and the development of village health services has extended the coverage to the vast majority of the Gambian population, urban and rural. Using immunization as one simple index, a 1996 survey found that 76.7% of all infants were fully immunized as judged by immunization cards and 87% of all 1-year-olds were fully immunized using the card plus the mothers' reports ( Central Statistics Department/UNICEF 1997; Tables 4.1–4.4). A 1990 household survey indicated that 92% of mothers delivering in the 5 years before the survey had received a tetanus toxoid immunization and that 68.8% of mothers had been assisted at the delivery by a nurse/midwife or by a trained birth attendant ( Pacqué-Margolis et al. 1993 ; Tables 7.1 and 7.2). These are remarkable achievements for a country with a per capita income estimated at US$ 320 in 1995 ( IBRD 1997)( Table 1) and health expenditures estimated at US$ 2.20 per head ( Gambia/UNICEF 1997).
Table 3. Economic characteristics of PHC and non-PHC villages in 1996
Table 1. Primary Health Care resources in the Kerewan Local Government Area 1997
Materials and methods
The study area for test of the effectiveness of PHC comprises 40 villages near the town of Farafenni on the north bank of the river Gambia, approximately 200 km inland from the capital, Banjul ( Figure 1). Fifteen of the villages have been designated PHC villages. The study area excludes villages in which the use of drugs, especially antimalarials, was common in 1981. These larger settlements were avoided because one of the original aims was to study the effectiveness of interventions against malaria, principally chemoprophylaxis and bednets. For the same reason, the study area also excludes the town of Farafenni (1993 population 21 000 people) and the area 10 km around the town. Most of the study area villages lie in a narrow strip of land on either side of the North Bank road no further than 22 km to the west and 32 km to the east of Farafenni.
Geographically, the Farafenni area is flat Sudan savannah with wet season cultivation of rice, millet, sorghum and groundnuts. The climate is sub-Sahelian with a short rainy season during the months of June through October. In the period for which rainfall data has been recorded at the MRC field station in Farafenni (1982–96), rainfall averaged 656 mm annually (S.D. = 135 mm). At the start of the study in 1981, medical facilities in the area were limited. The study population was initially served by three dispensaries, one in Farafenni town staffed by a dispenser and a midwife, and the other two in large villages west and east of Farafenni. Transportation to and from Farafenni was limited to bicycles, horse and donkey carts and some bush taxis travelling on dirt roads ( Greenwood et al. 1990b ).
In 1983, a new health centre opened in Farafenni town staffed by two to four doctors. Regular bus services were begun between the larger villages and Farafenni on the North Bank road in 1986 and ambulance services between the two dispensaries and Farafenni health centre became available around the same time. Telephone service was also added to the larger villages in the early 1990s. A donor-supported programme for the provision of modern pump wells began in 1986–87. In 1996, 88% of compounds had a reliable supply of safe water ( Hill et al. 1996 ). Most compounds have pit latrines.
In the vicinity of the study area, there are four larger ‘key’ villages (Illiassa, No Kunda, Ngeyen Sanjal and Sara Kunda) which have full-time CHNs ( Figure 1). There are dis- pensaries in Njaba Kunda and Ngeyen Sanjal. Eight villages (No Kunda, Illiassa, Sara Kunda, Dippa Kunda Wollof, Bambali, Katchang, Palen Wollof and Yallal) have trekking stations visited monthly mostly for antenatal care and vaccinations. There are 15 PHC villages and 25 non-PHC villages in the Farafenni study area. In addition a district health centre in the town of Farafenni was upgraded to a hospital in 1998. Table 1 gives the PHC resources in the Kerewan Local Government Area, comprising two health districts, in 1997.
The census at the beginning of the study in 1981 gave a population of 12313. Ethnically, 48% were Mandinka, 30% Wolof, 16% Fula and 6% other ( Greenwood et al. 1990a ). The study population on 31 March 1996 (the census years run from 1 April to 31 March the subsequent year) was 16416, with 45% of the population Mandinka, 34% Wolof and 20.3% Fula ( Table 2). Over the period 1981–96, the population has grown at 2.1% per annum which is much lower than the natural increase rate of 2.7%, indicating the important role played by out-migration. Almost half of the population is under age 15.
Table 2. Demographic characteristics of the village populations in the study area
Villagers living in the study villages are subsistence farmers. Few adults have been to school ( Table 3). The population is poor; in 1996 less than 10% of compounds owned bicycles, two-thirds of compounds possessed a radio, 35% owned some form of cart, and less than half slept on iron or wooden beds ( Hill et al. 1996 ). A 1992 survey revealed that the total fertility rate was 7.5 births per woman, mean birth intervals were 33 months and just 9% of women were using contraception, traditional or modern. Just over 3% of women had attended school and 56% of women aged 15–19 were married. Polygyny is common with 51% of women with one or more co-wives.
The PHC and non-PHC villages are located within a few kilometres of each other ( Figure 1) and so share similar ecological conditions. The PHC villages are larger with on average 719 residents, whereas the non-PHC villages have on average just 227 residents. Otherwise, the villages are quite similar ( Table 3). The number of residents per compound is a little larger in the PHC villages mainly because compound size differs by ethnicity. Mandinka and Wolof have larger compounds (20.6 and 21.6 people, respectively) than Fula (15.6 people). We have looked at possible interactions between ethnicity and mortality in all the villages and can find no significant effects for this population.
Despite the differences in village size and ethnicity, it seems there are few major differences between the PHC and non-PHC villages in income, education and wealth, although the exceptions seem to be the proportion of homes with a corrugated iron roof and the proportions of women using modern contraception ( Table 3). The contraceptive use rates are very low in both sets of villages, however, and seem to have little effect on fertility ( Table 3). There are no major development projects that are restricted to one set of villages or the other. For example, the well-drilling programme, as shown by the similar percentages of tube wells ( Table 3), covered villages of all sizes. More important for our purposes, we see little difference in the vaccination coverage for children in the two sets of villages ( Table 3).
Since the paper is based on short-term changes in the ratios of child mortality in PHC vs. non-PHC villages, formal testing for interaction effects between village size, ethnicity, possession of iron roofs or the use of modern contraception and childhood mortality seemed unwarranted.
Description of the data collection and analysis
The mortality estimates presented in this paper are derived from the continuing registration of residents, migrants, births and deaths over the period 1982–96. The principle source is the Farafenni Demographic Surveillance System (FDSS), a set of files compiled by continuous updating of the baseline information by itinerant field workers and resident village recorders. Additional and independent surveys complement the information base and serve as a check on the quality of routinely collected data. Data have been collected using similar procedures since inception although computers replaced the older ledger-based system used in the first few years of surveillance.
As with all surveillance systems, especially for small areas, the major difficulty is quantifying the flow of in and out-migrants as well as those moving within the study area. In addition, except during surveys of malaria chemoprophylaxis during pregnancy in some of the PHC and non-PHC villages ( Greenwood et al. 1989 ; Menendez et al. 1994 ), the system did not record and monitor the course of pregnancy until April 1998. For these reasons, periodic re-surveys are needed to update the basic census information and to recover some of the omitted vital events. Just after such re-surveys, recently conducted every two years, or following the conduct of a special survey, the information held on file inevitably changes. This makes the calculation of a consistent set of rates a difficult task since changes are made to the current ‘open’ census files which often have an effect on the older data. To dampen the oscillating trends in the numbers due to the data collection procedures and to deal with some of the interannual fluctuations in births and deaths due to varying rainfall amounts as well as to small numbers, we have combined the data into periods which we think correspond best to the level of development of the PHC system in the area ( Table 4).
Table 4. Periods used for the analysis of the effects of PHC in the study area
PHC was not introduced uniformly at a single moment in the district but instead progressed in a series of stages depending on local initiatives. On the North Bank, it took time for villages to identify TBAs and Village Health Workers to be sent for training. Although the PHC program was announced in 1983, it took a year or two before all 15 PHC villages in the study area were fully operational. We have taken the preintervention and the start-up period together to be 1982–86. The next period, 1989–96, corresponds with the time when Gambia's economy was growing strongly (despite the 1983–5 drought) and extra resources were available for the new PHC initiatives. Donor support for the programme was also strong throughout this period. Between 1992 and 1996, economic difficulties and political changes led to a stalling of the economy and to reduced supervision of the PHC system by the Medical and Health Department.
The first period for which we have mortality rates is the census year 1982–83 before PHC was established. The subsequent period 1983–86 is the first post-PHC period. These two periods were used by Greenwood et al. (1990b) in an earlier paper on childhood mortality in the district. It is impossible to reconstruct some of these early data since Greenwood et al. (1990b) used some PHC villages for trials of Maloprim to prevent malaria illness and death ( Allen et al. 1990 ). In these studies, individual children within half of the selected PHC villages were randomised to Maloprim treatment or placebo. The identity of these children is unknown today so we cannot now reconstruct these early rates excluding the children given Maloprim during the intervention. Nonetheless, we do have Greenwood's published rates for mortality in the PHC and non-PHC villages for the first two periods in which he excluded the children in the PHC villages that had received the Maloprim treatment. These rates, corrected for the effects of this intervention, could be compared with the rate calculated subsequently. We have no record of any other major intervention trials over the period under consideration that might otherwise bias the comparison of childhood mortality in PHC and non-PHC villages. From 1989 onwards, the record keeping of the surveillance system was being computerized and updated but the same procedures (i.e. village recorders, itinerant field workers and annual or biennial censuses) were used for data collection since inception. From 1989, however, we do have censuses ‘frozen’ at the end of the census year which allow us to reconstruct more fully the person-years of exposure for narrow age groups and time periods. Period, not cohort, death rates have been calculated using similar methods throughout the entire surveillance period so we believe the rates for the different periods can be compared directly.
The quality of the basic data has been improving steadily as more and more of the vital events fall within the period of surveillance beginning in 1981–82. Since 88% of children received at least the BCG vaccination in the 1990s, written records of the dates of birth of young children are with increasing frequency obtainable on health cards. Deaths are relatively easy to identify by surveillance and through the re-enumerations conducted recently every two years. The person-years at risk are more difficult to estimate due to migration between study villages and between study villages and other areas. The place of residence of the child often differs from the original site of enumeration, but for children under 10 we find that less than 1% of moves are between PHC and non-PHC villages. Most moves are within the same village.
One of the difficulties in assessing the contribution of the PHC system to infant and childhood mortality improvements is the strongly seasonal nature of childhood death rates. Since 1992, in almost every year, 60% or more of deaths of all 1–4 year-olds have occurred in the second half of the year which encompasses the short rainy season (normally June to October) when malaria transmission is most widespread. In addition, annual rainfall amounts fluctuate considerably with consequent effects on malaria deaths in particular. Since we have data on month and year of death for both infants and 1–4-year-olds, we are able to ascertain whether or not there has been a change in the distribution of death by calendar month over the surveillance period and whether or not there is a difference between PHC and non-PHC villages.
Smaller non-PHC villages generally surround PHC villages ( Figure 1) and so it is likely that some residents of non-PHC villages made use of the better services available at the PHC level. Indeed, mothers are used to walking to the key villages for antenatal visits and childhood vaccination provided by the trekking teams that visit from the district health centre on a regular rota. Other things being equal, this pattern of use of higher-order services by non-PHC village residents will tend to bias for PHC/non-PHC village comparisons towards zero.
The numerical results are summarized in Table 5 and in Figures 2 and 3. Both infant and under-five mortality rates fell dramatically from the high levels first recorded for the early 1980s, but there have been more recent reversals in the overall trend. Nationally, it seems that some 35% of children died before age 5 in the period before the 1970 census, falling to about 10% in the early 1990s ( MacLeod et al. 1998a ).
Table 5. Infant and childhood mortality rates for PHC and non-PHC villages 1982–96
Rate ratios and attached 95% confidence intervals indicate that infant mortality rates were lower in the PHC villages for the first 4 periods but the differences are not statistically significant at the conventional level. In the most recent period (1994–96), note that the infant mortality rate was higher in the PHC villages, but further analysis shows that the trend in the rate ratios over the whole period is not significantly different from one period to another.
Rate ratios for the mortality of 1–4 year-olds indicate that the rates were very similar for the first two periods. In the third and fourth periods (1989–94), rates in PHC villages are lower than in non-PHC villages. The differences are statistically significant for the third period and nearly significant at the 95% level for the fourth period. For the final period, the differences between the PHC and the non-PHC villages disappear. These changes are seen more clearly in Figure 3.
Turning to the deaths of infants by calendar month shown in Figure 4, we see a very slight tendency for death to be less concentrated in the rainy season in PHC villages than in non-PHC villages. For 1–4 year-olds, we see less concentration of deaths in the rainy season for some years (1991 and 1992, for example) but this change is mostly related to rainfall amounts and distribution. In 1991 and 1992, rainfall amounts in Farafenni were just 515 and 500 mm, respectively. There appear to be no major differences in the concentration of 1–4-year-olds deaths in the rainy season between PHC and non-PHC villages ( Figure 5).
Over the 1982–96 period, we find that in a poor, rural and largely uneducated population ( Table 3), substantial gains were achieved in the survival of young children in both PHC and non-PHC villages. Many factors are involved in this secular improvement in child survival in The Gambia ( MacLeod et al. 1998a, b ). The data support the hypothesis that PHC contributed a small additional part to the difference in the survival of 1–4 year-old children in PHC villages when the village-level health services were strongly supported by the government, donor agencies and the villagers. The lack of impact on infant mortality is notable.
We have already mentioned the likely effects of residents of non-PHC villages using the services of the PHC villages and how this would tend to reduce the effects of the PHC/non-PHC village comparison. Although the literature describes several interventions conducted in these villages, we can find none that might affect the differences in mortality between the PHC and the non-PHC villages. The well-known double intervention trial of Maloprim and impregnated bednets, tried in PHC and non-PHC villages in 1988–91, was conducted on the south bank east of Soma (see Figure 1), in a separate set of villages some 40 km away from those used in this analysis ( Alonso et al. 1993 ).
The importance of the results reported here is that using good longitudinal data for a rural area, we are able to exclude many of the variables that commonly confound the isolation of the independent effect of the health services on child survival. Such confounding variables include rising educational and income levels as well as modern family planning services. Very few socio-economic changes are recorded in all the study villages and none that singled out PHC or non-PHC villages for special treatment.
Some disaggregation of the under-five mortality rates by age provides some clues to the kind of factors implicated in the general decline of early age mortality. Neonatal deaths constitute a large proportion of all infant deaths (41% in the study area for the 1982–96 period) so we have studied the trend in neonatal mortality separately. There is a steady decline from rates of around 60/1000 at the beginning of the period to around 20/1000 at the end of the period with large interannual variations. For both neonatal and postneonatal death rates, we see little significant difference between the trends for PHC and non-PHC villages. In terms of the health interventions offered for this age group, we note that as part of antenatal care The Gambia offered tetanus toxoid immunization. This intervention is thought to have made an important contribution to the reduction of neonatal mortality in many developing countries during the 1980s ( Boerma & Stroh 1993). The care provided by trained TBAs also includes improved delivery practices and care of the umbilical stump ( Greenwood et al. 1990a ).
Some of the changes in the rates for under-ones may also be attributed to the malaria chemoprophylaxis programme for primigravidae seen at antenatal care. This intervention is known to be effective in the reduction of low birth weight ( Greenwood et al. 1989 ; Menendez et al. 1994 ). The slightly lower figures for neonatal mortality in the PHC villages compared with the non-PHC villages might be partly attributable to this effect, although most mothers seeking tetanus toxoid immunization congregate in the key villages visited by the trekking teams, regardless of whether they live in a PHC or a non-PHC village.
For children aged 1–11 months, mortality rates have changed quite slowly. Some of the lack of major improvement can probably be attributed to the persistence of undernutrition and ignorance of the importance of oral rehydration therapy to deal with severe dehydrating diarrhoea ( Gambia/UNICEF 1997).
For older children, the achievements have been more substantial ( Figures 2 and 3 and Table 5). The important feature is the mortality differential between PHC and non-PHC villages. Health interventions available in both that affect the mortality of 1–4-year-old children include promotion of improved health-related behaviour within the household by CHNs and VHWs; community-based public health measures such as the introduction of insecticide-treated bednets, vector control, and sanitation; easier access to some basic care at the village level, e.g. treatment of malaria at village level ( Greenwood et al. 1988 ); and the promotion of the use of oral rehydration therapy for acute diarrhoea. There was also some general health education in PHC villages that was not extended to the non-PHC villages.
Malaria chemoprophylaxis was given under supervision by village health workers to half the children in selected PHC villages in the period 1983–87 whilst other children in the same villages received a placebo ( Greenwood et al. 1988 ; Allen et al. 1990 ; Menon 1990). Adding these villages into the calculations of the mortality rates for PHC villages after 1989 might have reduced the mortality rates there. We think this effect is very slight since the prophylaxis stopped at the end of the intervention trial and the lasting effect of the drug treatment is small. Since there is very little transmission of malaria in the dry season, families do not buy chloroquine or Maloprim outside the rainy season.
A large part of the overall decline in child mortality over the last 10 years has been attributed to the Extended Programme for Immunization (EPI) in children ( Desgrées du Loûet al. 1996 ). Measles in particular was a major cause of mortality in The Gambia before the introduction of the immunization programme at the end of the 1970s ( McGregor 1991). Vaccination coverage for the Farafenni area in 1982 was already at the same level as at present, and has been fairly constant throughout the study period ( Greenwood et al. 1987 , 1990b). Hence, the overall decline in child mortality in both PHC and non-PHC villages from 1982 onwards in the study area is unlikely to be related for a major part to EPI. It is likely that factors such as improved health services, improvements in transport, and a general increase in health awareness have played a role in bringing about these changes.
A recent claim has been made for a dramatic increase in child mortality in Senegal due to chloroquine-resistant malaria ( Trape et al. 1998 ). This is not borne out by our data, by data from the Bassé area in eastern Gambia (Jaffar & Greenwood unpublished observation) or information from the ORSTOM/IRD surveillance site at Niakhar in Senegal, just 100 km from the Farafenni study villages ( Marra et al. 1995 ; Delaunay 1998). Although chloroquine-resistant malaria was detected in the Farafenni area in 1986, and has increased progressively since then ( Müller et al. 1996 ), there are no signs as yet in this study of either a significant sustained increase in child mortality or a changing ratio of rainy season to dry season deaths ( Figures 4 and 5). In any case, the difference between the PHC and the non-PHC villages is likely to be small over the period, since chloroquine usage was not extensive even in PHC villages (M. Pinder, personal communication).
Our analysis is a cautious endorsement of the value of adopting the PHC approach to the provision of basic health services in rural areas. This is the first report to show a significant additional effect of PHC on the survival of 1–4 year-olds. The effects on the survival of infants are less clear cut. Over the same period, we known that national mortality levels were improving steadily, even beyond the Farafenni Study Area ( MacLeod et al. 1998a ; Hill et al. 1999 ). When supervision of the village-level health services was vigorous and extra resources were directed to PHC, the mortality differentials between PHC and non-PHC villages widened.
In recent World Bank publications there is a shift in attention from community-based health care at village level to strengthening basic health services at higher levels ( World Bank 1993, 1994). This is in part a reaction to discouraging assessments of the contributions of village health care systems to health improvements. In many cases, the apparent lack of success may be partly due to poor functioning of dispensaries, health centres and district hospitals to support the village health care initiatives, the lack of supervision and continuing education of VHWs and TBAs rather than to the system itself. The results from The Gambia are more encouraging.
It is becoming clear that major improvements in child survival in rural sub-Saharan Africa can occur without major improvements in living standards or educational levels. The improvement in The Gambia, the fastest in sub-Saharan Africa for the 1988–96 period ( UNICEF 1998), took place while vaccination coverage was high at the start and remained fairly constant throughout the observation period. Improved basic health services, improvements in transport, urbanization and a general increase in health awareness probably have played a role in bringing about these changes. An important additional differential effect was observed in the survival of children in PHC villages when extra resources were devoted to village-level health services. Principally these effects seem to have been achieved through the promotion of improved health-related behaviour within the household by CHNs and VHWs; community-based public health measures; and easier access to some basic treatment at the village level. There was also some general health education in PHC villages that was not extended to non-PHC villages. The work here and subsequent analyses in process provide evidence of the added value of low-cost village health services in resource-poor populations.
The authors thank the many people who over many years contributed to the establishment and development of the surveillance system at Farafenni. They include the successive Heads of Station (Andrew Bradley, Robert Snow, Arun Menon, Steve Allen, Pedro Alonso, Steven Lindsay, Umberto d'Alessandro and Olaf Müller). Others who have been helpful in different ways include Reuben Mboge, Fatou Banja, Greg Fegan, the field workers Louie Loppy, Tumani Trawalla, and Birran Samba, and the many interviewers who have worked on the series of studies conducted in the study area. The continued interest of the former Director of MRC (Brian Greenwood) and the current Director (Keith McAdam) has been important in maintenance of the system. Brian Greenwood and Abdoulie Jack are acknowledged for their comments on an earlier draft of this paper.