The Safe Motherhood Model is a policy tool designed for use by those whose work is to apportion the resources available for medical care coverage provided to women during three states of reproduction: ante-natal care, delivery care, postpartum care. We concede that maternal issues are only part of the spectrum of reproductive health issues, but there is more than enough here to challenge both the builders and users of this model. The primary source of data for the model is the PERFORM study. As the model evolved, numerous assumptions were introduced, and simplifications made. These were done to strengthen it as a prospective policy tool at the expense of it being a retrospective statistical study. This does not imply dissatisfaction nor criticism of the PERFORM study. When questions arose, PERFORM was compared with NFHS 1992-93. We have no reason to believe that PERFORM does not reliably reflect conditions in Uttar Pradesh.
We do have reason to believe that the maternal health care delivery system does not meet the needs of women in Uttar Pradesh. The basis for this conclusion is outlined below. In short, maternal health care coverage cannot be correlated to a reduction in morbidity (and we presume, to mortality either). In fact in most instances there is an increase in morbidity among covered women. This anomaly can, in part, be explained by 1) increased awareness of covered women, 2) an assumption that women, anticipating problems, would seek coverage in greater proportion than their counterparts, and 3) the propensity to seek coverage may be related to other social and economic factors confounding the results. We do not believe that any amount of statistical analysis can confidently reverse the significance of these outcomes.
Thus we have a dilemma. A clear need for maternal health services can be documented. Existing health care services cannot be shown to reduce morbidity. Increasing delivery of existing health care services cannot be expected to reduce morbidity.
The only solution to this dilemma is to assume that existing services can be improved to achieve some significant reduction in morbidity (and mortality) and coverage extended to those in need of those improved services.
PERFORM (Program Evaluation Review and Organizational Resource Management) was devised to monitor and evaluate improvements in access, quality and demand for family planning and reproductive health services in Uttar Pradesh (UP). The baseline (1995) survey was designed to provide insights into the current state of family planning services in U.P. Twenty-eight districts, two from each of the fourteen divisions of the state, were surveyed. The Safe Motherhood Model was based on data gathered from the Women's Schedule, Section 8, Reproductive Health. This section provides the maternal history for the eligible population. The eligible population is defined as married women, aged 14 to 49.
NFHS (National Family Health Survey 1992-93) is an implementation of the Demographic and Health Surveys for India. The sample from Uttar Pradesh is comparable to that of PERFORM in many instances.
The ability to relate inputs to outcomes is essential to the task of the policy maker. One needs reasonable assurance that social investments and expenditures achieve positive outcomes. The two following examples relate antenatal care to antenatal and partum outcomes.
High coverage is distinguished by the client receiving a relatively complete set of antenatal services. Clients classified as low coverage receive at least one service, but not the threshold defining high coverage. Problem outcomes are distinguished by the reporting of one or more negative conditions of some significance (detailed definitions are included in model documentation).
|
Antenatal Coverage |
Problem |
Normal |
Total |
Problem |
Normal |
Coverage |
|
High |
2,197 |
1,592 |
3,788 |
57.99% |
42.01% |
15.47% |
|
Low |
5,184 |
5,024 |
10,208 |
50.78% |
49.22% |
41.68% |
|
None |
4,629 |
5,865 |
10,494 |
44.11% |
55.89% |
42.85% |
|
Total |
12,010 |
12,481 |
24,491 |
49.04% |
50.96% |
100.00% |
Note that the incidence of problematic conditions increases with level of coverage. Some would argue this is simply a matter of women with problems seeking coverage. That is, problems cause coverage instead of coverage causing problems. The problem with that explanation is that a woman with four antenatal visits is covered before the problems occur.
We can avoid the time issue entirely by looking at antenatal coverage and partum outcomes such as deliveries.
|
Antenatal Coverage |
Problem |
No problem |
|||
|
Number |
Percent |
Number |
Percent |
Total |
|
|
High |
1489 |
39.31% |
2299 |
60.69% |
3788 |
|
Low |
3468 |
33.97% |
6740 |
66.03% |
10208 |
|
None |
2875 |
27.40% |
7619 |
72.60% |
10494 |
|
Total |
7832 |
31.98% |
16658 |
68.02% |
24490 |
In this case we get the same correlation between level of coverage and incidence of problems.
A common response has been "It can't be!" Indeed the survey could be honed to provide sharper indications of both inputs and outcomes. But a similar pattern exists in the NFHS study.
|
Antenatal Coverage |
Proportion of live births |
|||||||
|
Caesarian |
Forceps |
Bleeding |
Labor |
Placenta |
Breech |
Other |
Total |
|
|
High |
4.9% |
1.6% |
4.5% |
6.6% |
1.4% |
0.3% |
1.2% |
20.5% |
|
Low |
4.0% |
0.2% |
2.4% |
6.0% |
0.8% |
0.1% |
0.8% |
14.3% |
|
None |
1.0% |
0.1% |
3.0% |
5.8% |
1.0% |
0.0% |
0.9% |
11.8% |
High coverage is 4+ visits, low coverage 1-3. The definition of problem is more specific and less inclusive than PERFORM. This data does not include sepsis. The correlation between level of coverage and incidence of problems is similar. It should be noted that this pattern exists in India as a whole, and even in Kerala where maternal care is thought to be better.
In the face of demonstrated need that 50% of women have negative outcomes in the antenatal period and over 30% in delivery period (PERFORM), this conclusion is cynical.
Since there is no evidence that maternal care reduces the incidence of problem outcomes, we could argue that the degree and duration of disability from those outcomes is reduced by treatment. This data is not available in either study. In this model we have resorted to "expert opinion" to assign relative weights to treated and untreated problem conditions.
There exists some optimism that protocols could be adopted which would reduce the incidence of problems. Speculation about such change, even when done by "experts", presupposes profound changes in the delivery system. Without multiple providers, it is difficult to imagine such change occurring bureaucratically from the top. Nevertheless, we have sought "expert opinion" on what "might" be accomplished given the requisite political will.
We accept the overall incidence of problems from the PERFORM study.
We assume that case management reduces both the degree and duration of disability.
We assume further, granting improvements in maternal care delivery, that reductions in incidence can be achieved.
The PERFORM questionnaire directs the interviewer to skip questions about outcomes of pregnancy if "anything was done to end the pregnancy prematurely". In many cases, this directive was ignored and the interviewer proceeded to ask about outcomes such as sepsis or hemorrhage. The use of such data is thus called into question.
The proportion of terminations is inconsistent with mortality attributed to informal (or uncovered) terminations. Estimates of termination range from 10 to 25% while PERFORM reports slightly over 1%.
Women who say the pregnancy was not wanted are three times more likely to miscarry than women who say the pregnancy was wanted.
To do this we would have to adjust the distribution of deaths by partum to reduce case fatality rates for non-covered terminations to a credible level.
Accept estimates of covered and non-covered terminations from "expert sources".
We use the lower end of the range of estimates of terminations. This preserves a significant role for terminations in the calculations while provoking less discussion on the incidence of termination.
We assume some of the miscarriages of women who did not want this pregnancy were actually terminations
Wantedness is presumed to have some effect on the distribution of partums. In PERFORM we find wantedness has little effect on terminations but a larger effect on miscarriages. This anomaly may be the result of miscarriage being a more socially acceptable explanation for the termination of a pregnancy than an induced termination.
The reporting problem is actually quite small in comparison to the difference in ex post and ex ante indications of wantedness. Only 12.4% (NFHS 92) of currently married women want another child soon. Some of these women are undoubtedly non-susceptible (12.4% NFHS 92) and 13.1% sterilized. On the other hand, for 84% of the maternal histories (PERFORM), women said that pregnancy was wanted.
Then there are matters of degree. Since the model attempts to relate wantedness to decisions affecting partum distribution, we would prefer an indicator consistent with a woman's decision to consider an induced termination.
Initially we made the assumption that pregnancy during use was unwanted and pregnancy after discontinuation was wanted. Further, some proportion of pregnancies of non-users were unwanted as well. An outcry over the use of continuation without simultaneously looking at the effects of acceptance led to removing all reference to continuation in the family planning submodel (this is a topic in itself).
The current version assumes a proportion of wanted pregnancies by method. Clearly this is 0 for sterilization, somewhat more for IUDs and somewhat more again for pills. An estimate of wantedness for nonusers is added. In combination with prevalence and effectiveness, an overall proportion of wantedness is calculated.
We encountered two problems in estimating effectiveness from PERFORM data. Prevalence is in general low but for specific methods, injectables, IUD, and pills, the sample size is too small to be reliable. For example, there were 6 users of injectables one year before interview. One of them became pregnant in the year preceding the interview. As a class, injectable users are more likely to become pregnant than nonusers. As a result injectable effectiveness is negative.
The second problem is that effectiveness by method in Uttar Pradesh is unacceptably low.
|
Method |
Prevalence |
Effectiveness |
|
None |
80.2% |
0.0% |
|
Pill |
1.0% |
12.8% |
|
IUD |
1.1% |
53.3% |
|
Injectables |
0.1% |
-8.1% |
|
Condoms |
3.2% |
46.1% |
|
F. Ster. |
11.7% |
98.3% |
|
M. Ster. |
1.4% |
84.4% |
|
Other |
1.3% |
18.2% |
|
Total |
100.0% |
15.1% |
Following a similar chain of reasoning to maternal care effectiveness above, we have used values closer to those observed in Bangladesh. Using PERFORM values, suppliers would be motivated to discontinue injectables and pills. In view of low ex ante wantedness and low contraceptive prevalence, such a policy is inadvisable.
In the model, morbidity and mortality are combined as disability adjusted life years (DALY). The application of the DALY is much simpler than that proposed by Murray, et al (The Global Burden of Disease, Harvard 1996). Specifically, the model uses 30 years as the burden of a maternal death. This is largely consistent with the age weighted and discounted factor used by Murray (a death at 30 years of age = 29.92 years). The burden of a maternal "problem" is assigned directly but is based on a simple rectangular function of degree of disability times duration.
Several estimates of maternal mortality exist for Uttar Pradesh and the mean figure of 600 per 100,000 live births was adopted. This number is reasonably consistent with the indirect estimation from NFHS.
The deaths were then distributed across antenatal, partum and postnatal periods. Initially we used the distribution of deaths from a hospital study in Lucknow. We were later persuaded to use a distribution from Matlab in Bangladesh. The largest difference was a shift from antenatal to partum. The partums were divided into miscarriages, terminations, premature and normal deliveries. From the number of cases and the number of deaths, we derived mean case fatality rates by stage and type of partum. Since PERFORM (and all surveys) interviewed surviving women, there was no basis for estimating case fatality rates from the survey.
The efficacy of case management on case fatality rates was obtained from "expert opinion". Although surveys can and do include opinions of widowers on whether the death of a woman was pregnancy related, the generally accepted notion of maternal death depends on more authoritative opinion.
No cost data was available from either PERFORM or NFHS.
It is not sufficient to identify the more cost effective interventions and increase their coverage. More of the same in most instances would not serve the interests of "Safe Motherhood".
There is ample evidence that the lack of effectiveness is systemic in Uttar Pradesh. There are no obvious changes in operations (protocols) or administration that would dramatically improve the outcomes. Thus, an increase in the variety of delivery systems may be required for there to be significant progress in this area.
The agreement between PERFORM and NFHS suggests that the data collected is not in grievous error. In general, more information about inputs (including costs) and outcomes (especially degree and duration of disability) are in order. It should be noted that PERFORM included a maternal history and contraceptive use calendar and NFHS did not. A calendar is essential to estimating contraceptive effectiveness.
Coordination between survey and vital statistics would improve handling of mortality.
Many aspects of the PERFORM study remain, incidence of problem outcomes (morbidity), coverage of maternal health interventions, and contraceptive prevalence. Some variables such as mortality never came from PERFORM. Measures of efficacy and effectiveness from PERFORM are negative. Policy makers cannot make sensible judgements in the face of negative outcomes.