Measures

A number of different measures of maternal mortality are commonly used, the most important being the maternal mortality ratio and the maternal mortality rate. The numerator for both is the same: the number of maternal deaths occurring in a given period. Most studies adhere to the International Classification of Disease, ninth revision (ICD-9), definition of maternal death in including the deaths of women within 42 days of termination of pregnancy. Some older studies may include deaths that occurred up to 90 days after the termination of pregnancy, in accordance with the Guide of Maternal Deaths Studies published by the American Medical Association (1964).

The maternal mortality ratio is the ratio of maternal deaths per live births (variously 1,000, 10,000, or 100,000; unless noted otherwise a metric of 1,000 is used for ratios and rates in this paper). Although the denominator should ideally be pregnancies, the impossibility of obtaining accurate counts of fetal losses and stillbirths has necessitated using live births. According to the World Health Organization (WHO) (1985), in countries with low induced-abortion rates, the number of live births is within 10 percent of the number of pregnancies. This ratio measures the probability of maternal death, the obstetric risk.

The maternal mortality rate is simply a cause-specific death rate: maternal deaths/women of reproductive age (variously 10 to 49, 15 to 44, 15 to 49 years old). This rate is a function of the incidence of pregnancy as well as the risk inherent in pregnancy. Thus, the maternal mortality rate is linked to the maternal mortality ratio through the general fertility rate (births/1,000 women of reproductive age).

Two extensions of the maternal mortality rate have been used as measures of maternal mortality: (1) the proportion of all-cause mortality for women of reproductive age that is attributable to maternal mortality and (2) the lifetime risk of maternal mortality, calculated as reproductive-span maternal mortality rate (Measham and Herz, cited in Fortney, 1987).

It is important to consider the effects of age and fertility structure differences on these measures. Obstetric risk varies greatly with age, as does the incidence of pregnancy. Thus, populations that have different age structures and/or different age-specific fertility rates may have different crude maternal mortality ratios even when they have similar age-specific ratios (see Fortney, 1987, and Graham and Airey, 1988, for examples). Whether or not the crude ratios are different depends on whether the ages of highest obstetric risk are those in which the population or fertility differences reside. The crude maternal mortality rate and the measures based on it are even more sensitive to these structural differences.

Unfortunately, nearly all studies report only crude maternal mortality ratios and rates, making comparisons across populations or even of the same population at successive times questionable.

Sources of Data

Maternal mortality is underestimated even in countries with excellent vital registration systems; official statistics from countries where maternal mortality is high seriously underestimate the true level (WHO, 1987). Better data are available from three sources: population-based studies, hospital population studies, and case series. Population-based studies provide the least biased estimates of maternal mortality. They allow nearly complete counts of live births and maternal deaths. Because of their prospective nature they include deaths that occur more than a week after delivery and probably include most of the deaths due to induced abortion. There are very few population-based studies concerning maternal mortality: six from Bangladesh (Chen et al., 1975; Lindpaintner et al.,1982; Khan et al., 1986a; Alauddin, 1987; Koenig et al., 1988; Faveau et al., 1988); one from Ethiopia (Kwast et al., 1986); one from Egypt (Fortney et al., 1985);¹ one from the Gambia (Greenwood et al., 1987); and one from Jamaica (Walker et al., 1985). Preliminary results from a population-based study in India are available (Bhatia, 1985). Although the Medical Research Council (MRC) project in the Gambia and the Machakos project are population laboratories that have yielded excellent information on a number of topics, maternal mortality in both areas has been almost completely eliminated because of care provided by the project (Lamb et al., 1984; Voorhoeve et al., 1979), and so these studies are not included. In addition, one hospital-based study from Lusaka, Zambia (Mhango et al., 1986), arguably covers a sufficient portion of the population (85 percent of births, 90 percent of deaths) to be considered a population-based study. Unfortunately, the number of deaths in each study is generally very small, so estimates by specific cofactors are unstable.

With regard to morbidity the WHO collaborative studies on family formation patterns and health provide population-based information about anemia and uterovaginal prolapse in nine countries (Omran et al., 1981a, 1981b). In addition, one study from Jerusalem (Harlap et al., 1971) concerns obstetric interventions, and one from the Philippines (Raymundo, 1987) concerns morbidity during the last pregnancy.

Hospital population studies, particularly those from referral hospitals, provide less accurate estimates of the incidence of maternal mortality or morbidity than do population-based studies. They reflect the experience of only a proportion of the population during only part of the risk period at the time of and immediately after pregnancy outcome. Many of the hospital-based studies report only the crude maternal mortality ratio and limit further analyses to the series of deaths. A few report case-fatality rates for various complications. If the assumption can be made that the deaths in hospitals are representative of all maternal deaths or the way in which they are unrepresentative can be identified, then some examination of the proportional contribution of various causes becomes worthwhile.

Case series—reports about a series of deaths, either due to any maternal cause or to a specific cause, such as ruptured uterus, eclampsia, or hepatitis—can provide valuable information about incidence and case-fatality rates. Unfortunately, most contain no description of the population from which the cases are drawn. However, series selected because of the presence of a risk factor (e.g., the Arkutu, 1978, series of primigravidae) rather than an outcome and those in which the maternal deaths are compared to a subsample of the hospitalized population represent one of the most efficient ways of obtaining information about maternal mortality.

Problems of Measurement

As noted above, the maternal mortality ratio as defined by the WHO is already an approximation of the measure of interest. However, the more serious problems in measurement arise from incomplete ascertainment of either the number of deaths or the number of related live births. Deaths that occur before delivery, such as those due to ectopic pregnancy, may not be recognized as maternal deaths. Additionally, those due to induced abortion may be misrepresented because of shame or fear of prosecution: in one series of cases of women who came to a hospital with tetanus ''all [22] postabortal cases followed criminal abortion, although in only 10 cases was the interference admitted by patients or their relations'' (Adadevoh and Akinla, 1970). Another group of deaths that is easily missed, especially in hospital-based studies, are those that occur some time after delivery. The population-based study in Tangail (Alauddin, 1987) shows 15 percent and that in Jamalpur (Khan et al., 1986a) shows 27 percent of deaths occurring more than a week after delivery.

Depending on the time period of interest, retrospective population-based studies may underestimate both maternal deaths and live births. A valid count of live births is generally less of a problem for prospective population-based studies but exists nevertheless. Comparison of the crude birth rate for the Jamalpur prospective study with the national rate suggests an underreporting of up to 10 percent of the births (Khan et al., 1986a); the same comparison for the Tangail prospective study suggests underreporting of 27 percent (Alauddin, 1987).

The problem of estimation of the denominator is most crucial, however, with regard to hospital-based studies. In developing countries many women do not usually deliver in hospitals, although they may be brought there if the delivery is complicated. Hospital-based maternal mortality ratios in urban areas with easy access are probably overestimates because a larger proportion of more difficult than normal deliveries is represented among deliveries. This tendency for hospital populations to overrepresent abnormal or complicated cases is exacerbated in the case of referral or teaching hospitals in large cities, which receive not only all the self-referred emergency deliveries but also patients referred by other health facilities. As most hospital reports are generated by teaching or referral hospitals, the reported maternal mortality ratios are likely to be overestimates. One study reported two ratios, the first 16.7, calculated on the basis of hospital births, the second, 3.9, on the basis of births in the city (Rao, 1975).² Unfortunately, the number of births in the hospital catchment population often cannot be ascertained or estimated; this might happen, for example, if there are several hospitals in a city. The representativeness of the hospital population can be assessed using two observations that are frequently reported: the proportion of admissions that are emergency or unbooked and the proportion of deaths that occur soon after admission.

In areas where access to a hospital is restricted (because of distance, cost, or social barriers), a greater proportion of women experiencing difficulty during delivery will die at home. Ratios reported by these hospitals may represent underestimates; in addition to the advantage of medical care at the time of delivery, the population delivering in hospitals may have had greater than average exposure to antenatal care and may represent a more socioeconomically advantaged portion of the population. One extreme example is from the Medical Research Council (MRC) study area in the Gambia: the maternal mortality ratio for the period 1951–1975 was reported as about 10 (Billewicz and McGregor, 1981), but during 8 years after the establishment of a continuous medical service there were no maternal deaths, although 16 would be expected if the established ratio had prevailed (Lamb et al., 1984). An additional problem of smaller rural facilities is that caseloads are often insufficient to produce stable estimates of mortality.

While estimates of maternal mortality ratios from hospital-based studies may be biased, they provide a useful supplement to the few estimates obtained from population-based studies. However, because of the very small numbers of deaths identified in population-based studies,³ using hospital-based studies is crucial for examination of the causal mechanisms involved in high maternal mortality.

If all the maternal deaths from an area occur in a hospital, the proportion of mortality associated with each cause of interest can be determined accurately (apart from classification bias), even when the maternal mortality ratio cannot be determined. This situation probably occurs more frequently when a large proportion of the population is already delivering in a hospital—for example, in Lusaka (Mhango et al., 1986). When it does not occur, three types of deaths are most likely to be excluded. First, many early-pregnancy deaths occur at home: in Addis Ababa 6 of 13 abortion-associated deaths occurred at home. Some hospital studies explicitly exclude all early-pregnancy deaths (e.g., Chi et al., 1981; Hartfield, 1980) as well as those occurring after the woman has been discharged from the hospital, while a few simply do not report any (Balachandran cited in Armon, 1977; D'Cruz et al., 1986a).

The second type of death apt to be omitted from hospital statistics comprises those due to indirect causes—not resulting from the complications of pregnancy itself but rather from a condition that is aggravated by pregnancy or one that is completely unrelated, such as a motor vehicle accident. Kwast et al. (1986) report hospital deaths for 67 percent of those dying of obstetric causes but only 38 percent of those dying of indirect causes. Even if they die in hospitals, women who die of indirect causes may, for example, die in the medical ward and never come to the attention of those who report "maternal mortality." Finally, women are more apt to be seen in a hospital if the delivery complication "affords relatives the time to discuss the merits of hospital admission" (Hartfield, 1980) and the hospital is sufficiently close to the woman's home. One way to estimate this effect is to examine the ratio of deaths from obstructed labor and hemorrhage to those from hypertensive disease; this ratio tends to be much higher in hospital-based studies than in population-based ones.

In addition to these types of inclusion bias, hospital series reports are also subject to classification bias. For example, deaths resulting from cephalopelvic disproportion and abnormal presentations leading to prolonged labor and uterine rupture might be classified as due to ruptured uterus if the rupture is identified, as due to hemorrhage if extravaginal bleeding is a prominent feature, or to sepsis if death is delayed some time after the onset of labor. Many maternal deaths are associated with a number of complications; for example, of 219 deaths in Zaria only 86 could be attributed to a single condition, while the rest were associated with two or more (Harrison and Rossiter, 1985).

This then is the situation: maternal mortality is inherently difficult to measure; the summary measures commonly used obscure any differences between populations that might be due to fertility patterns as opposed to other sources of risk; the best (though still flawed) data come from a few extremely small studies in restricted geographic areas; and the largest source of data is subject to selection and classification biases that cannot easily be controlled for because they are unquantifiable except in the very particular circumstances of each study. Even with these drawbacks, fairly clear patterns are apparent both in terms of the geography of level and causal components and with reference to the relation between fertility and maternal mortality.

Parity/Gravidity

The event that puts a woman at risk of maternal mortality is conception. Strictly, the relationship between fertility and maternal mortality should be described in terms of pregnancies. Some risks (e.g., hemorrhage from ruptured ectopic pregnancy, complications of induced abortion, amplification of the risk of infectious diseases) accrue to pregnancy well before delivery, while others are inherent in delivery itself. Most of the studies concerning maternal mortality report parity-specific mortality; a few report rates by gravidity.⁴

All the population-based studies indicate and results from hospital studies generally confirm that the first birth and births of high order are strong risk factors for maternal mortality (Table 2). These studies indicate a J-or U-shaped risk with parity: high during the first pregnancy, lowest during the second or third, and high again by the fifth pregnancy. A similar pattern is found for gravidity, with an even stronger relative risk for first pregnancies relative to later-order pregnancies than is observed for first births.

Table 2

Mortality Per 1,000 Live Births by Parity and Gravidity Prior to Current Pregnancy.

Corroboration for the higher risk of first deliveries comes from the population-based morbidity study carried out in West Jerusalem, where the obstetric intervention rate was calculated by parity (Harlap et al., 1971). Obstetric interventions included any specialist intervention at any stage of labor, including surgical or medical inductions of labor, forceps or vacuum deliveries, breech deliveries, cesarean sections, and other major third-stage interventions; while probably an overestimate, this measurement does indicate potential mortality in the absence of medical care. Both the parity-specific rates and the ratios (adjusted for age, ethnicity, and hospital) indicate about a 70 percent higher risk for first than for second births (Table 3).

Table 3

Obstetric Interventions by Parity in Jerusalem.

Except for very high parity births (10+), this study does not confirm the excess risk of multiparity. This is surprising in light of the very clear enhanced risk of malpresentation for multiparous women noted by, for example, Faundes et al. (1974): 3.4 breech presentations per 1,000 women of parity 0, 10.1 for parities 1 and 2, 14.4 for parities 3 and 4, 17.5 for parities 5 and 6, and 19.0 for parity 7+. A review of a series of 50,057 deliveries, including 5,785 to grand multiparous (parity 7+) women in Haifa, Israel (Fuchs et al., 1985), yielded a rate of malpresentation of 11.9 per 1,000 for the grand multiparous women in contrast to a rate of 3.3 per 1,000 for women of lower parity.

A study from the Sudan (Aziz, 1980) comparing 2,049 primiparous, 3,679 multiparous, and 3,130 grand multiparous (5+) women found slightly higher rates of abnormal presentation for grand multiparas (14 percent compared to 9 percent for multiparas and 9.5 percent for primiparas). This same study found more toxemia among primiparas than other women (10.8 percent vs. 7.8 percent for multiparas and 8.5 percent for grand multiparas), more anemia among grand multiparas (7.4 percent) than other women (4.5 percent), and a clear increase in the risk of antepartum hemorrhage with parity (primiparas 1.3 percent, multiparas 1.9 percent, grand multiparas 4.5 percent). AI-Sayegh and Hathout (1974) in Kuwait obtained similar results for antepartum hemorrhage (0.1 percent for women of parities 1 to 5, 1.2 percent for parities 6 and 7, and 4.2 percent for parity 8+), but they observed lower rates of abnormal presentation—only 5.1 percent for women of parity 8 or more. Their case series is, however, quite a bit smaller, including 2,060 women of parities 1 to 5, 494 of parities 6 and 7, and 446 of parity 8+.

In a 1958–1960 multicenter study in the United States, Israel and Blazar (1965) reported a much higher rate of essential hypertension among grand multiparas (7+) (without controlling either for age or race), higher rates of uterine rupture and postpartum hemorrhage, and significantly higher rates of placenta previa and placental abruption, but no difference in maternal mortality, presumably because of adequate hospital care.

Interaction Between Age and Gravidity/Parity

Most information about complication or mortality is couched in terms of either age or parity: women are at greatest risk at young and old ages and at low and high parities. Because age and parity are strongly associated, it is not clear if the age-specific and parity-specific patterns reflect the same basic age-driven risk, if they have independent effects, or if they act in combination. Contraception provides the means of affecting timing of fertility and thus increases the importance of knowing whether there is an interaction between age and parity in their effects on morbidity and mortality associated with fertility.

Four of the population-based studies provide information about maternal mortality by age and parity. The simple J-or U-shaped relationship between parity and maternal mortality is less clear-cut when age is taken into account. Part of the reason for this is the small number of deaths in any of the studies; as a result, the age-parity-specific estimates are fairly unstable, even when some categories are grouped. Two of the studies (Chen et al., 1975; Khan et al., 1986a) provide sufficient data to permit the calculation of 95 percent binomial confidence limits for each estimate.

The Matlab data (Chen et al., 1975) show an overall U-shaped distribution: the lowest-parity women in the youngest age group and the highest-parity women in the oldest age group have the highest mortality. Women in the middle ages (20 to 39) and middle parities (2 to 5) have risks of mortality that do not seem very sensitive to increments of age or parity. While this may be an artifact of the way the data are grouped, it nevertheless suggests that younger primiparas and older grand multiparas are the women at greatest risk.

The pattern for Jamalpur (Khan et al., 1986a) is quite different, J-shaped rather than U-shaped. There is no apparent excess risk associated with a first pregnancy, which is rather surprising. This may be due to missed maternal deaths. The authors note that the crude birth rate calculated from the number of births reported in the study is 35, about 10 percent lower than the national estimated crude birth rate of 39. If births have been missed, it seems possible that maternal deaths might also have been. One social factor that would make it more likely that missed births would be first births rather than higher-order births is the custom of women going to their mother's home (and therefore away from the study area) for several months at the time of the first delivery. The increasing risk of mortality with parity in the middle age group contrasts with the rather flat pattern in Matlab.

The data for the third Bangladesh study, in Tangail (Alauddin, 1987), have been very reduced; all ages and parities have been grouped into nine cells, two of which are empty. The overall pattern is J-shaped, with very little difference apparent between 20-to 29-year-old women and those 30 or more.

The population-based study from Jamaica (Walker et al., 1985), with finer age and parity classifications than any of the studies from Bangladesh, shows the expected pattern of higher risk in the first than in the second pregnancy for all women less than 30 (Table 4). For 20 to 24 year olds and 25 to 29 year olds this pattern of decreasing risk with increasing parity continues through the fourth pregnancy, then increases sharply. For older age groups the lowest-risk pregnancy is at higher parities, suggesting that women of less than average fertility for their age may be less healthy. However, it is important to note that even at its lowest, mortality is about four times higher for these age groups than for the lowest-mortality group—the 20 to 24 year olds having their fourth child.

Table 4

Age-parity Specific Maternal Mortality Ratios in Jamaica.

One hospital-based study, from 12 hospitals in Indonesia, also provides mortality ratios by age and parity (Chi et al., 1981). The level of mortality is about half that in the studies from Bangladesh, but the same U-shaped pattern is apparent.

While evidence from these five studies is suggestive rather than conclusive, it seems that ages below 20 and above 30 enhance the simple parity-specific patterns already observed. This impression can be tested by examining several hospital series concerning primigravid/primiparous or grand multiparous women that also take age into account in describing complication rates.

Hospital Studies: Primigravid/Primiparous Women

Age-specific maternal mortality ratios for primigravidas in Zaria, Nigeria (Harrison and Rossiter, 1985) show that 20 to 24 year olds had the lowest mortality (4.8/1,000 deliveries). Below age 20, the risk of mortality varied inversely with age. Primigravidas with a stated age of less than 15 and those 15 years old had about the same very high mortality—38.5 and 41.9, respectively. Sixteen-year-old primigravidas had a mortality ratio of 14.0, and for those 17 to 19, it was 8.0. This marked pattern of higher risk with younger age is undoubtedly exaggerated by selection bias, as there is a strong local belief that home delivery is best for young girls. Primigravidas age 25 to 29 had a slightly increased risk relative to 20 to 24 year olds (6.1); those 30 or older had an even higher risk (10.1).

Among 2,291 primigravidas in Tanzania, women less than 20 years old had higher rates of anemia, eclampsia, premature labor, and mortality than women 20 to 24 years old (Table 5; Arkutu, 1978). A review of the records of women less than 20 years old who delivered at Kenyatta National Hospital in Nairobi in 1978 (Ngoka and Mati, 1980) indicated a slightly higher rate of premature delivery, with about 40 percent of the 13 to 14 year olds, 30 percent of the 15 to 16 year olds, and 20 percent of the 17 to 19 year olds delivering before 37 weeks of gestation. No mortality was observed in young primigravidas, compared with a ratio of 3.5 for the rest of the referral hospital population.

Table 5

Complication Rates and Death Rate Per 1,000 First Deliveries.

A study from Lagos (Efiong and Banjoko, 1975) that compared very young primigravidas (less than 16 years old) who had adequate or inadequate prenatal care with 22-year-old primigravidas who had adequate care indicated a higher rate of severe pre-eclampsia and eclampsia among the teenagers, even among those who had prenatal care. The authors speculate that this might be related to the nutritional status of the teenagers, who were on average from poorer families than the older primigravidas. Interestingly, the teenagers had lower rates of anemia (prenatal care, 22 percent; no care, 35 percent) than the older women (45 percent).

A similar study carried out in Jamaica from 1966 through 1969 (Hay and Boyd, 1973) found no difference in rates of pre-eclampsia or anemia but did report higher rates of premature rupture of the membranes (2.4 percent vs. 1.8 percent), antepartum hemorrhage (3.0 percent vs. 0.9 percent), prolonged labor (11.3 percent vs. 6.6 percent), and cesarean section (15.9 percent vs. 12.5 percent) among the older women than among the teenagers. They remark that most women, whatever their age, attend prenatal clinics regularly.

In Chile the prevalence of hypertensive syndrome increased with age at all parities, but within each age group it decreased with parity up to parity 7 (Faundes et al., 1974). Women having their first births had about double the risk of women having their second or third births at all ages except below the age of 20, when the risk for first births was three times that of second births. Primigravidas below 20 years of age were also at increased risk for hemorrhage of placental origin.

Except for the study in Jamaica, these hospital studies confirm the general pattern suggested by the population-based studies: young primigravidas are at higher risk of complications, especially pregnancy-related hypertensive disease, and of mortality than primigravidas age 20 to 24.

Hospital Studies: Multigravid/Multiparous Women

In Nigeria the age of lowest risk for multigravid and grand multiparous women (5+) was again 20 to 24 years old (Harrison and Rossiter, 1985). The mortality for all multigravid women that age was 5.9 per 1,000 deliveries; for 20-to 24-year-old women of at least parity 5 it was 9.7. Risk increased with age for both groups; for all multigravid women age 30 or more mortality was 15.3, while for the grand multiparous subgroup it was 20.1.

In Chile Faundes et al. (1974) found that for hemorrhage associated with uterine inertia the major risk accrues to women of parity 5 or more, and except for women above 40 years of age this pattern is basically unaffected by age. For hemorrhage of placental origin the pattern is similar in that most women 35 or more, and most women of parity 5 or more are at greater risk. The combination of age greater than 35 and parity greater than 5 does not greatly increase risk over that for the age or parity group. As reported above, this study showed the risk of malpresentation increasing with age; at the youngest ages the risk is the same for all parities, but at all ages above 19 the risk is at least twice as high for second and third as for first births.

A study carried out in the United States in 1958–1960 reported the incidence of anemia among younger and older, white and nonwhite women of high and lower parity (Israel and Blazar, 1965). Among white women they found three times more anemia (hemoglobin, <10 grams percent) in grand multiparas (7+) than in women of lower parity. The excess was the same for women less than 35 (32 percent vs. 10 percent) as for those 35 or more (24 percent vs. 8 percent). Among nonwhite women the prevalence of anemia among grand multiparas was about the same as for white women in both age groups, but there was a much higher prevalence of anemia among lower-parity women. Thus, for nonwhite women there was at most a 20 percent excess of anemia associated with grand multiparity (32.6 percent vs. 27.4 percent for women under 34; 26.5 percent vs. 24 percent for 35+). Interestingly, in all four race-parity groups, older women were less anemic than younger women.

While these studies suggest that the risks for most complications and for death increase with both age and parity, the relationship between parity and risk is more important. Within age groups the risk of mortality or complications generally increases with parity, but within parity groups age often has no additional effect.

Short Birth Intervals

No study yet identified specifically addresses the relationship between birth-interval length and maternal mortality in developing countries. From a hospital population study in Zaria, Nigeria, Harrison and Rossiter (1985) reported that for women of the same parity the risk increases as a factor of the proportion of previous children dead. On average, women with a higher proportion of previous children dead would, in Nigeria, be expected to have shorter birth intervals; however, it is possible to postulate an exogenous factor that would account for high levels of both child and maternal mortality.

A referral hospital population study carried out from 1936 to 1943 in the United States (Eastman, 1944) showed no significant difference in maternal mortality ratios by length of interval since the last live birth. The overall maternal mortality ratio was 3.1; the interval-specific ratios were 0 (0/115 deliveries) for intervals less than 12 months, 2.2 (3/1,347) for 13 to 24 months, 3.2 (7/2,191) for 25 to 48 months, and 4.0 (6/1,506) for intervals longer than 48 months. This study also provides information about some types of morbidity: there was no significant difference in the incidence of postpartum hemorrhage or puerpural fever by interval nor was any interval effect on the incidence of anemia evident. The incidence of toxemia was positively related to the length of the preceding interval (Table 6).

Table 6

Relationship Between Interval Since the Last Live Birth and Toxemia at the Time of Current Delivery.

Because the hospital in which this study was carried out was a referral hospital and the patients are thus unrepresentative of the overall population, the possibility must be considered that this relationship is spurious. However, there is no reason to believe that the length of interval was a criterion in referral. Moreover, toxemia is but one of a number of conditions for which patients were referred, and none of the others show any relationship to interval. A more plausible alternative explanation for this relationship is the tendency for intervals to be longer for older women, in conjunction with the increased prevalence of primary hypertension with age.

Whether the prevalence of anemia increases with parity is the subject of controversy. The population-based studies of Omran and Standley et al. (1981a, 1981b) provide information about mean hemoglobin levels by age and parity for women from nine countries. For India and rural Turkey there is a clear inverse relationship between hemoglobin level and parity, and there is some indication of such a relationship in the data from Lebanon and Pakistan, but the information from Iran, the Philippines, Colombia, Egypt, and Syria shows no such trend.

Lack of Antenatal Care

One of the most striking correlates of maternal mortality is lack of antenatal care. The excess risk for women who have not attended an antenatal clinic can be observed by comparing maternal mortality ratios for women who attended at least once with those for women who never attended (Table 7). Except in the Gambia, women who never attended died at 2.6 to 22 times the rate of clinic attenders. While surprisingly high, the rates of antenatal clinic attendance among all patients seem representative, at least of hospitals that report this statistic in Africa and south South Asia. Among the others in the series considered for this review, reported attendance rates were 40 percent in Saudi Arabia (Chattopadhyay et al., 1986); a national rate of 85 percent in Zambia (Mhango et al., 1986); 85 percent in Harare, Zimbabwe (Frost, 1980); and 90 percent in the Kilimanjaro region of Tanzania (Armon, 1979). Reported rates of ever attendance for eight African hospitals in the maternity care monitoring program ranged from 76 percent in Bamako, Mali, to nearly 100 percent in Mwanza, Tanzania (Janowitz et al., 1984), and recent statistics compiled by Royston and Ferguson (1985) indicate rates of similar magnitude.

Table 7

Mortality Among Patients With and Without Antenatal Care.

Hartfield (1980:71) suggests that the lower mortality among antenatal clinic attenders ''is due more to a greater awareness of the benefits of hospital care and a concomitant willingness to use it than to any specific antenatal therapy.'' This effect would help to explain the striking reduction in mortality among even the irregular clinic attenders in rural India (Shah et al., 1971). From the Zaria data it is also clear that attendance at an antenatal clinic is strongly related to access; among women delivering in a hospital, 81 percent lived in the same town as the hospital attended, while only 17 percent of those who came from outside the town had attended (Harrison and Rossiter, 1985). Thus, when the clinic is at the hospital (as was the case in India), having had antenatal care probably proxies for living fairly close to the hospital. In Zaria it was also strongly associated with education: of those delivering in the hospital, 97.5 percent of women with any formal education had attended the antenatal clinic at least once, while only 62 percent of those without education had done so. In all but one of the eight hospitals reporting about antenatal care in the maternity care monitoring program, the median number of antenatal visits increased with the women's level of educational attainment (Janowitz et al., 1984). Thus, it seems likely that at least part of the beneficial effect of antenatal care is due to its correlation with access to a hospital (both physical and psychological) and with better socioeconomic status.

However, some portion of the beneficial effect occurs because antenatal clinics help prevent some complications. The advantage of regular antenatal care is that women at particular risk (e.g., because of anemia, hypertensive disease, or an obviously small pelvis) can be identified and either treated to reduce the risk or advised to deliver in a hospital. Data from Lusaka, Zambia, corroborate this effect: women who had antenatal care had considerably less risk of dying of hypertensive disease than women who did not have such care, while the rates for hemorrhage and amniotic embolism deaths are about the same in the two groups (Table 8; Mhango et al., 1986). In South Africa Melrose (1984) also reports a striking proportion of unbooked patients—83 percent—among those dying of hypertensive disease, with the proportion among puerperal sepsis deaths being about the same.

Although mortality is much lower among women who attend antenatal clinics, assessment of the quality of care at one clinic in a district hospital in Kenya (Malone, 1980) revealed deficiencies. Even after procedures were changed to facilitate identification of women at risk of complications, only 60 percent of those with abnormal conditions and 84 percent of those who required hospital delivery were correctly referred. Moreover, not all of those who are referred follow advice. One study in Nigeria reports the deaths of four women who attended antenatal clinics and who were advised to have repeat cesarean sections but who defaulted until late in obstructed labor because they wished to avoid operative delivery (Caffrey, 1979). Similar behavior has been reported from Mkar, Nigeria (Groen, 1974), and it seems likely that it occurs elsewhere in Africa.

Distance

Up to 15 percent of pregnancies will involve complications, most of which are not preventable (Rosenfield and Maine, 1985). The major factor in mortality for these cases becomes delay in arriving at the hospital. The Zaria data provide some insights into how distance affects mortality. Patients living outside Zaria had higher mortality than those from the town, but 80 percent of all deaths occurred in women who lived less than 2 kilometers from a main road. This suggests that access was a major factor in determining who got to the hospital. The excess risk was greatly reduced among women who had received antenatal care (Table 9).

Table 9

The Effect of Distance and Prenatal Care on Mortality.

This information can be used to set upper and lower limits for an estimate of the true population ratio. Since 81 percent of those living in town had antenatal care, the overall town ratio is probably lower than the true population ratio. The overall ratio for those living outside the town is probably inflated relative to the true ratio, as 83 percent of those women were emergency patients.

The pattern of people who live closer to the hospital having lower mortality is confirmed by observations in Anantapur, India, where maternal mortality in the urban area is 5.5, while in the rural area it is 8.7 (Bhatia, 1985). At all ages a much higher proportion of urban than rural deaths occurred in hospitals (Table 10).

Table 10

Percent of Deaths Occurring in Hospital by Age and Urban/Rural Sector of Anantapur District, India.

Socioeconomic Status

In Ethiopia Kwast et al. (1986) found that when abortion deaths were excluded, illiterate women had the highest mortality (4.8 vs. 3.3); when they were included, illiterate women still suffered the most mortality, (7.1), but the rate for women with 7 or more years of education was nearly as high (6.0), while that for women with only a little education was lower (4.5; difference not statistically significant). Women from households with a monthly income of less than $25 had significantly higher mortality than women from households with a higher monthly income (15.3 vs. 3.2).

Counterintuitively, Alauddin (1987) reports for Tangail that maternal mortality is higher among women from more solvent families and among those whose families hold 2.01 to 3 acres of land (MMR = 10.3, compared with 5.3 for the women whose families hold the least land and 5.8 for the richest). Similarly, mortality is higher for those who have primary school education (14.3) than for those with no education (3.2) or secondary schooling (8.7). While some of this differential is due to higher abortion-related death rates among literate and primary-level educated women than among illiterate and higher-educated women (presumably because the rate of abortion seeking is higher in these groups), the basic pattern persists when all abortion-related deaths are excluded. Alauddin (1987) points out that the 1981–1982 Nutrition Survey of Rural Bangladesh found an inverse correlation between income and the prevalence of nutritional anemia among pregnant and lactating women and also found that women in families with intermediate-size landholdings were at highest risk of anemia.

Traditional Behavior

Differences by Area

Even a cursory examination of the proportional mortality for the population-based studies (Table 11) reveals some striking patterns: abortion is important in nearly every area, responsible for about 10 percent of the deaths in Jamaica, 20 percent in Bangladesh, nearly that proportion in Zambia, and nearly 30 percent in Ethiopia. Hemorrhage is a major cause of death in every study, and hypertensive disease is also important in most areas. Zambia, Ethiopia, and Gambia have high proportions of deaths attributed to infectious disease, which is responsible for very few deaths in Bangladesh, Jamaica, and India.

Table 11

Proportional Mortality—Population-Based Studies.

The hospital studies from West Africa (Table 12) show very few deaths due to abortion. Overall, hemorrhage is the most important cause, with difficult labor second, followed by infectious disease, and finally hypertensive disease. The studies from East and South Africa (Table 13) show a much larger proportion of deaths due to abortion, and hemorrhage seems less important than in West Africa. Infectious disease accounts for at least 10 percent of deaths. Hypertensive disease shows an interesting pattern, accounting for two to four times as many deaths in the series from Zambia and South Africa than in Uganda and Tanzania. Deaths from thromboembolism are also more prominent in South Africa and Zambia than in East Africa. They are virtually absent in West Africa.

Table 12

Proportional Mortality in Hospital Studies in Nigeria.

Table 13

Proportional Mortality in Hospital Studies in Africa, Excluding West Africa.

Abortion accounts for about 10 percent to 20 percent of deaths in the studies from India, except for the rural area near Bombay (Table 14). Hemorrhage is again the most important single category, but hypertensive disease accounts for at least 10 percent of deaths. The proportions of deaths attributed to anemia are consistently much higher than in the East and South African studies and in most cases are higher than the proportions from West Africa. Virtually all of the infectious disease deaths are attributed to hepatitis, in contrast to the pattern in Africa, especially West Africa, and in the Middle East.

Table 14

Proportional Mortality—Hospital Studies, India.

The Middle Eastern pattern (Table 15) is quite different from those in India and Africa: neither difficult labor nor hypertensive disease is very important, and anemia is totally absent. Abortion is responsible for 10 percent to 20 percent of deaths, and hemorrhage is fairly important. What is particularly striking is that the studies from Iran have larger proportions of ''other infectious'' deaths than any other studies, and those from Lebanon and Saudi Arabia have the highest proportions of deaths from thromboembolism.

Table 15

Proportional Mortality in Hospital Studies in the Middle East, Indonesia, and Vietnam.

In the Indonesian studies hemorrhage and sepsis are the most important causes and possibly hypertensive disease. The high maternal mortality ratio and the pattern of important causes from the hospital in Vietnam—abortion, difficult labor, and operative—suggests that it was really a last resort.

Abortion

Abortion accounts for a substantial proportion of maternal mortality in all regions covered except West Africa. The proportion of these deaths attributed by the authors to induced abortion varies. Some authors state (Walker et al., 1985) or imply (Barford and Parkes, 1977; Melrose, 1984) that all were induced. Lindpaintner et al. (1982), who made a special effort to identify induced-abortion-related deaths, report that of 10 abortion-related deaths, 1 abortion was spontaneous, 6 were reported to have been induced, and 3 were suspected to have been induced. Similar statements are common in the studies from Africa; and the study from Riyadh states that all 3 identified abortion-related deaths were assumed to be related to spontaneous abortion (Chattopadhyay et al., 1986). Thus, it is not possible to determine maternal mortality related to unwanted fertility. However, most of the evidence suggests that most abortion-related deaths are in fact due to induced abortion.

Parity-specific patterns are very different in Africa and Asia. The population-based studies from Bangladesh and hospital studies from India suggest that in Asia most deaths are to multiparous women. This pattern is confirmed by a population-based study of abortion in Jamalpur, Bangladesh, which found that the induced-abortion/live birth case ratio increased from 32/100 for primiparous women to 37 for parities 1 to 4 and 96 for women of parity 5 or more (Khan et al., 1986b). Moreover, case fatality also increased with parity, from about 1 percent for primigravidas to 3 percent for those of parity 5 or more. Most induced abortions were obtained by married women. Only 12 percent were obtained during the second month of pregnancy and 49 percent during the third.

A hospital-based study of women admitted for complications of induced abortion in Dhaka confirmed the high parity and married status of most women who obtain an abortion in Bangladesh (Khan et al., 1984). The hospital case-fatality rate was 7.7 percent; and compared with women who survived, the women who died were less likely to be married (55 percent vs. 91 percent), more likely to be slum dwellers (58 percent vs. 23 percent), more likely to have gone to a nonmedical practitioner (95 percent vs. 45 percent), and more likely to have had abortion induced by insertion of solid objects into the uterus (71 percent vs. 30 percent). The major complications were uterine perforation, excess bleeding, and sepsis.

Interviews with health workers throughout Bangladesh about pregnancy-related deaths in 1978–1979 revealed that an average of 25 percent of identified deaths were due to induced abortions, with estimates for specific districts ranging from 16 percent to 48 percent (Rochat et al., 1981). The estimated proportion of maternal deaths due to abortion in Tangail District was 33 percent, about twice the 17 percent obtained from the population-based study carried out 5 years later (Alauddin, 1987). The health worker interviews suggested a proportional mortality of 21 percent in Mymensingh District, the site of the Jamalpur population-based study, which also obtained a proportional mortality of 21 percent (Khan et al., 1986a).

The population-based study in Anantapur India (Bliatia, 1985) and the hospital-based studies from India suggest that proportional mortality of abortion is at about the same level or slightly lower than in Bangladesh. One hospital-based study of abortion in Thailand (Chaturachinda et al., 1981) suggested that women who have abortions there, like those in Bangladesh, tended to be older and to have more children. A survey of knowledge, attitude, and practice in one region of the Philippines in 1967–1969 also showed a similar pattern and found that the prevalence of ever having had an induced abortion was 30 percent (Flavier and Chen, 1980).

In contrast, 12 percent of Turkish women with only one or two living children said in response to a direct question that they had had an induced abortion; the proportion estimated by randomized response technique in a similar population was 30 percent for those with one or two children and 17 percent for those with none (Tezcan and Omran, 1981).

The pattern in Africa is similar to that in Turkey: Aggarwal and Mati (1980) found that 53 percent of women hospitalized with septic abortion in 1978 in Nairobi were less than 20 years old, compared with 35 percent of those with nonseptic abortions. Infection is used as a proxy for induced abortion, as women do not necessarily admit to having induced an abortion. Aggarwal and Mati state that most patients were primigravidas. Eighty-three percent were in their second trimester, and the mortality ratio was 2/1,000 abortion admissions (12.5/1,000 septic abortion admissions).

A study of 59 women in Nigeria who were admitted for complications of abortion and admitted to having induced the abortion showed that 52 percent were less than 18 years old (Okojie, 1976). For 81 percent it was the first pregnancy; most were single, and 58 percent were secondary school students, suggesting that abortion was being used to delay a first birth. Other studies from Nigeria support this conclusion (Unuigbe et al., 1988; Adetoro, 1989). This was also the conclusion of Lamptey et al. (1985), who surveyed women coming for delivery in a hospital in Accra, Ghana. They found that 25 percent reported having at least one induced abortion, and they mention that the case-fatality ratio in their hospital for induced abortions was at least 34 per 1,000.

Given this reported prevalence and extremely high case-fatality rate, it is surprising that abortion figures so little in the hospital-based studies of maternal mortality from this area. One possibility is that deaths are missed because women are admitted to a different section of the hospital; both the Aggarwal and Mati (1980) and the Okojie (1976) papers discuss induced or suspected induced abortions solely in the context of gynecologic emergency admissions. A second possibility is that behavior in the capital city—Nairobi or Accra—is different from that in more provincial cities and that abortion rates are higher. If the reported reluctance of women to admit having recently induced an abortion reflects a reluctance to admit it implicitly by seeking care, inclusion bias probably plays a stronger role in exaggerating case-fatality rates for abortion than for delivery mortality. Nevertheless, induced abortion resulting from unwanted fertility clearly increases maternal mortality.

The pattern of utilization of abortion by young primiparous women implies that it may be difficult to reduce maternal mortality from this cause by increasing contraceptive use. The chances for that are much greater in Asia, where not only are the women who seek abortion older but also, at least according to one series of interviews with abortion providers, their husbands agree with their decision (Islam, 1982).

Difficult Labor

Difficult labor is more important as a cause of maternal mortality in West Africa than anywhere else, although the studies from East Africa also show high proportions. A large number of these deaths are associated with uterine rupture.

A number of case series studies concern uterine rupture (Table 16). Incidence rates range from 0.7 to 10.8 per 1,000, no doubt at least partially due to variation in the prevalence of hospital delivery. What is most striking is the very high associated mortality: the lowest among the series is 7 percent and the highest is 42 percent, with a median rate for all studies of 10 percent. Much of this excess mortality is due to late arrival at the hospital; except for the series from South Africa, fewer than 20 percent of patients in any study had had any antenatal care, indicating that they arrived as emergency admissions.

Table 16

Uterine Rupture in Africa: Incidence, Case-Fatality, and Associated Risk Factors.

The distribution of type of rupture differed widely between studies. From 1 percent to 63 percent were attributed to rupture of a sear, usually a previous cesarean section, but occasionally a repaired rupture or tear and in a few cases sequelae of an induced abortion. Between 6 percent and 32 percent were attributed to trauma, mostly arising from manipulations during attempts to correct malpresentation. In all series except the one from Cameroon, most ruptures were considered to be spontaneous. More than 50 percent of these were associated with fetopelvic disproportion, in several cases exacerbated by traditional attempts to remedy the situation (use of oxytocin-containing medicine, abdominal pressure on uterus).

These hospital series of uterine rupture contain few primigravidas. The average parity for nearly all series was about 4, and risk apparently increases with parity. However, the proportion of women with rupture who are at less than parity 5 ranges from 42 percent to 83 percent (Table 16).

In a retrospective study of maternal height as a predictor of disproportion in primigravidas in Sierra Leone, almost 8 percent of a consecutive series of clinic attenders had cesarean section for disproportion (Aitken and Walls, 1986). While one criterion for inclusion in the study was a complete record, including height and weight, few records were rejected for incomplete data, and there is no suggestion that height was selectively recorded for those thought to be at risk of disproportion. If this is taken to be a representative population, the implication is that cesarean section averts some of the deaths that would otherwise occur because of disproportion and uterine rupture. Five of the hospital studies of maternal mortality include information about cesarean section rates and about associated case fatality (Table 17). Unfortunately, none of them report parity-specific rates or give much detail about the reasons for the cesareans, so it is difficult to assess the fertility-associated risk of disproportion. Aitken and Walls (1986) state that it is highest for primigravidas, but they offer no evidence.

Table 17

Incidence and Case-fatality of Cesarean Sections.

Studies in Latin American hospitals in Honduras, Venezuela, Chile, and Brazil show a much higher incidence of cesarean section, ranging from 5 percent in Honduras to 41 percent in Brazil (Janowitz et al., 1982a). Associated case-fatality rates ranged from 0 (Honduras and Brazil) to 0.3 percent (Venezuela); maternal mortality per 1,000 vaginal deliveries in the same hospitals ranged from 0 (Brazil) to 0.3 (Venezuela), much lower than in any of the hospital series from Africa or Asia. The age-parity-specific incidence rates showed the same pattern in all four hospitals: highest in primigravidas and decreasing with parity. Age above 30 years increased the risk of having a cesarean section at all parities but especially for primigravidas, with this multiplicative effect decreasing with parity. Across all hospitals the major recorded indication for primigravidas having a cesarean section was obstructed labor, but from 27 percent to 83 percent of cases had no recorded indication. A second study, of two hospitals in Brazil, suggested that recorded indications are not a good criterion of true risk; incidence of cesarean was most strongly related to method of payment (Janowitz et al., 1982b).

Even though the underlying causes for the observed inverse relationship of parity and risk of cesarean section cannot be precisely determined, the results of these studies point in the same direction as the information from the series about uterine rupture: a great deal of the risk of death from difficult labor occurs to women of lower parities. The potential for averting these deaths through reducing fertility may be limited.

Hemorrhage/Anemia

Hemorrhage is the most consistently important cause of death across all areas covered by the hospital studies. It is associated with a large number of conditions, including obstructed labor and placental pathology, which are not very susceptible to simple methods of detection or prevention. Late arrival at hospital is a major cofactor in hemorrhage deaths (Armon, 1979; Chi et al., 1981).

Although underlying anemia can be a cofactor in death from hemorrhage, observed associations between anemia and maternal mortality probably more often reflect the effect of hemorrhage and delay in hospitalization. Anemia measured from blood taken at admission reflects blood loss due to the complication as well as underlying anemia. Women who have lost more blood before arriving at the hospital are probably more likely to die; thus the relative risk of mortality associated with anemia is probably overestimated, and many "anemia" deaths are likely to be mainly due to hemorrhage.

Underlying anemia can exacerbate the effects of hemorrhage: with the same amount of blood loss, anemic women suffer from shock more quickly than nonanemic women (McFee, 1973). Most underlying anemia is due to nutritional deficiencies; usually lack of iron or folic acid. Some is due to malaria infection (see infectious disease section below) and a small amount is a consequence of hemoglobin SS or S-C disease.⁵ Anemia is most prevalent in Asia, particularly South Asia; followed by West Africa, then East and South Africa; with Latin America having the lowest rates (Table 18), and is more common in rural areas than in cities (Royston, 1982). It is strongly associated with poverty. In addition to contributing to deaths from hemorrhage, anemia is an important factor in cardiac failure.

Table 18

Estimated Percentage of Pregnant Women With Hemoglobin < 10 g/100 ml.

In the hospital series reviewed, anemia is cited as a factor in many of the deaths in Bangladesh (Alauddin, 1987), Tanzania (Armon, 1979) and Nigeria (Ojo and Savage, 1974), where it is noted as being most severe in primiparas. Chattopadhyay et al. (1986) report that the prevalence among the obstetric population in Riyadh is less than 1 percent. Only two studies cite hospital population prevalence and give case-fatality rates. In Madurai, India, the prevalence of severe anemia (< 7 grams percent) was 3.2 percent among all women coming for delivery; the maternal mortality ratio of severely anemic women was 45.2, compared with 15.8 for the rest of the population (Rao, 1975). In Indonesia, the prevalence of anemia (< 10 grams percent, measured on admission) was 47.7 percent for rural and 31.4 percent for urban women (Chi et al., 1981). The maternal mortality ratio was 13.5 for rural anemic women but only 2.3 for nonanemic women. For urban women the differential was less: 3.8 for anemic, compared with 1.8 for nonanemic women. The authors suggest that because of transportation difficulties rural women arrive later during complicated deliveries than urban women do and have thus suffered a greater loss of blood prior to admission.

In the hospital studies reviewed that discuss association with age or parity, hemorrhage is remarked to be a cause of death associated with high parity and older age: 75 percent of the hemorrhage deaths in Durban were to women more than 25 years old (Melrose, 1984), while in Pietermaritzburg most were above the age of 30 and of parity 5 or more (Barford and Parkes, 1977). In Jamaica the maternal mortality ratios for hemorrhage increase with age up to 8.1 for those 35 to 39 and 5.1 for those 40+ (Walker et al., 1985). In Chile most of the risk of hemorrhage, whether due to placental pathology or uterine inertia, accrued to women of parity 5 or more, except that primigravidas less than 20 were also at higher risk of placental hemorrhage (Faundes et al., 1974).

Thus, except for malaria-associated anemia (see below), any increase in risk of hemorrage death because of the prevalence of anemia is either stable or increases with parity. Moreover, there is evidence that hemorrhage occurs mostly in older multigravid women. Hemorrhage deaths, then, are one category of maternal mortality that might be reduced by a reduction in fertility.

Hypertensive Disease of Pregnancy

Virtually all of the studies reviewed, whether population based or hospital based, and whether concerned with mortality or morbidity indicated that hypertensive disease of pregnancy is most common among primigravidas. The pattern of increased risk for younger primigravidas below the age of 20 may reflect not so much increased physiologic risk as other differences between them and women who have their first child at age 20 to 24. In a recent prospective study of primigravidas in Vietnam, Burma, Thailand, and China (WHO International Collaborative Study, 1988), clinical diagnosis of hypertensive disorder varied greatly, from about 1 percent of the women in Vietnam and Thailand, to 5 percent of those in Burma and 31 percent of those in China. The proportions of women having different signs of hypertensive disease differed among the countries, but for all signs except proteinuria the proportion was highest in China and lowest in Vietnam. The most striking finding was that mean diastolic blood pressure was remarkably constant across populations through the second trimester; in the third trimester there was little increase in Vietnam or Burma, there were greater increases in Thailand, and the largest increase was in China. Whether due to genetic differences, as the authors suggest, or to subtle differences in behavior, it is clear that cross-national differences are significant.

Hypertensive disease noticed in older primigravidas and in a few studies among older grand multiparous women (e.g., Barford and Parkes, 1977) may reflect the increased risk of essential hypertension with age.

Infectious Disease

Although not many studies deal explicitly with this aspect of maternal mortality, it is clear that pregnancy increases the case-fatality rate of certain infectious diseases—infectious hepatitis, malaria in some circumstances, and possibly others—probably through alterations in the immune system in response to pregnancy. In tetanus-endemic areas, abortion and delivery offer excellent opportunities for the organism to infect women.

Hepatitis

In the population and hospital studies reviewed, infectious hepatitis is the single most important disease in terms of increasing the risk of maternal mortality. It is not clear whether pregnant women are more susceptible to hepatitis than nonpregnant women, but they are at greater risk of death from the disease than nonpregnant women. Case series reports demonstrate case-fatality rates that range from 10 percent to 61 percent (Table 19), much higher than for the general population. The series that deal exclusively with pregnant women have been criticized on the grounds that these rates might be spuriously high because only seriously ill women would be admitted to a hospital. Several reports compare hospital case-fatality rates for pregnant women with those for nonpregnant women and men; these confirm that pregnant women are at particular risk and also that women in general are at greater risk than men (Table 19).

Table 19

Case-fatality Rates per 100 Hospitalized Patients With Hepatitis.

There remain several questions concerning possible selection bias. Christie et al. (1976) considered whether pregnant women with jaundice might be more likely to be admitted than nonpregnant women with jaundice. This possibility was rejected; however, it is important to note that even if such a bias existed, it would tend to reduce the case-fatality rate among pregnant women. A more serious problem is whether women come later and are more severely ill on admission than men and whether this tendency is greater for pregnant than for nonpregnant women. Gelpi (1978–1979) rejected the first part of this hypothesis, finding no significant difference by sex in the proportion of fatal cases presenting in coma. In Ghana, Morrow et al. (1968) found no substantial difference in hospitalization rates by sex for patients from the Accra Municipal Council Area (for males, 30.9 per 100,000; for females, 34.8), but pregnant women had higher case-fatality rates and a higher risk of serious disease.

Haemmerli (1966) raised the question of genetic susceptibility, noting the predominance of high case-fatality rates in the circum-Mediterranean area; this seems unlikely given the additional case series reports from India, Ghana, and Ethiopia (Wahi and Arora, 1953; Morrow et al., 1968; Kwast and Stevens, 1987) as well as the high proportional mortality from hepatitis reported in some of the hospital maternal mortality series (D'Cruz and Balani, 1968b; Shah et al., 1971; Konar et al., 1980; Barford and Parkes, 1977; Ojo and Savage, 1974).

The generally higher case-fatality rate for nonpregnant women than for men has been attributed to the lower nutritional status of women, but no evidence has been offered to support this.