antenatal screening: its in assessing obstetric risk ...journalofepidemiology andcommunityhealth...

Journal of Epidemiology and Community Health 1994;48:297-305

Antenatal screening: its use in assessing obstetricrisk factors in Zimbabwe

Vivien D Tsu

AbstractStudy objective - To assess the predictiveutility of obstetric risk factors for identi-fying before the onset of labour thosewomen at high risk of obstetric compli-cations in a developing world setting,where home deliveries predominate andemergency transport is scarce.Design - Risk factors derived from twopopulation based, case-control studies(one of cephalopelvic disproportion andone of post partum haemorrhage), car-ried out in Zimbabwe were used to con-struct weighted and unweighted scores, avariety of screening algorithms, and setsof probabilities estimated from logisticregression models. These screening testswere evaluated for sensitivity, specificity,positive predictive value, and "cost" (theproportion of the population testing pos-itive). Each complication was evaluatedseparately and the two were then pooled.Participants - All were Harare residentswith singleton, vertex deliveries andspontaneous onset of labour. A total of201 experienced cephalopelvic dispro-portion, 150 had post partum haemor-rhage, and 299 had normal, unassisteddeliveries.Measurements and main results -Largely because of the very low incidenceof the two complications studied (1% orless), positive predictive values were low(less than 7%). Holding "cost" constantat 10%, a screening test for cephalopelvicdisproportion could predict 42-3% ofcases compared with only 35 0% of thosewith post partum haemorrhage.Weighted scores had little advantageover unweighted ones, and probabilitiesfrom the logistic regression models didnot differentiate cases from controls verywell.Conclusions - With simple algorithmsbased on maternal height, parity, andobstetric history, more than one third ofwomen at risk for potentially fatal com-plications could be identified at relativelysmall cost to themselves or the healthcare system.

(J Epidemiol Community Health 1994;48:297-305)

One of the primary goals of antenatal care is toidentify those women with raised risks forproblems during pregnancy or delivery, inorder to ensure that precautionary measuresare instituted where possible or more intensivemedical care is available should it be needed.

In the developing world, where most womendeliver outside formal health care facilities andemergency transport is often difficult or im-possible to obtain, the ability to identify thosewomen who are at high risk of delivery compli-cations and who ought, therefore, to deliver ata hospital would be very valuable. Such refer-rals could greatly reduce the present highlevels of maternal mortality and morbidity, ifwomen followed the recommendations and ifeffective hospital care were then provided.Despite all the attention paid to identifying

"high risk" women during pregnancy, thereare no published studies linking risk factoridentification with reduced maternal or peri-natal morbidity or mortality. There are manyreasons for the paucity of data on this complexinter-relationship, including lack of concep-tual clarity among clinicians working in thisarea, multifactorial outcomes and aetiologies,logistical difficulties in implementing neededresearch, and ethical dilemmas.

Unlike other types of epidemiological re-search which look at risk factors as a means ofuntangling aetiologies, obstetric risk factorsare often used as a screening device to assist inpatient management. In this situation, a riskfactor is noteworthy if it is an effective predic-tor of the outcome of interest, even if it is onlyindirectly related to the outcome and cannotitself be changed or prevented. Risk factoridentification is often focussed on secondaryrather than primary prevention, since manymaternal complications can be treated but notprevented from occurring.

Research into obstetric risk factors has notbeen a high priority in industrialised countries,since most women have ready access to compe-tent medical care there and poor maternaloutcomes are relatively rare. Maternal riskfactors have been studied primarily for theirvalue in predicting or preventing unfavourableperinatal outcomes.`15 Such research has beendifficult to undertake in developing world set-tings, where hospital based obstetric patientsare usually unrepresentative of parturientwomen generally,8 and data on deliveries out-side hospital are quite difficult to obtain. Theproblems are compounded by the diversity ofcomplications for which screening is employedand the inter-relatedness of the potential riskfactors. Different components within an out-come may have different risk factors that can-cel each other out when lumped together foranalysis. In addition, there is a considerabletime lag between the risk assessment and theoutcome of interest, during which conditionscan change and interventions can take place.Evaluation is complicated by the fact that thebenefits and hazards of the test are difficult to

Department ofEpidemiology,University ofWashington,SeattleVD TsuCorrespondence to:Dr V D Tsu,2532 11th Avenue West,Seattle, WA 98119,USA.

Accepted for publicationSeptember 1993

297

on October 16, 2020 by guest. P

rotected by copyright.http://jech.bm

j.com/

J Epidem

iol Com

munity H

ealth: first published as 10.1136/jech.48.3.297 on 1 June 1994. Dow

nloaded from

http://jech.bmj.com/

298

quantify as are the consequences of beingincorrectly classified.The present study was undertaken in an

effort to predict risk for two well definedcomplications for which the interventions havealready been proved relatively effective, postpartum haemorrhage and cephalopelvic dis-proportion. A site was chosen (Harare, Zim-babwe) that permitted nearly 90% of all localbirths to be represented in the study, while stillbeing able to include data on a wide range ofpertinent variables, data that were collectedbefore the outcome was known.

MethodsRisk factors for the two complications wereidentified through a multivariate logisticregression analysis of data from Harare, Zim-babwe. The data came from two case-controlstudies involving 203 women with cephalopel-vic disproportion, 151 women with post par-tum haemorrhage, and 299 women with nor-mal vaginal deliveries. Details on the subjectsand data collection and analysis are reportedelsewhere.78 Briefly, all cases and controls werewomen resident in Harare with singleton, ver-tex deliveries and spontaneous onset of labour.Cephalopelvic disproportion cases includedeligible women delivering in Harare duringeight months in 1989 with an operative deliv-ery for cephalopelvic disproportion. In addi-tion to meeting a strict set of criteria designedto ensure that cephalopelvic disproportion wasthe primary cause of the operative delivery,eligible cases were rated as "definite" or"probable" by at least two of a panel of threesenior obstetricians who independentlyreviewed the abstracted case records. Post par-tum haemorrhage cases occurred in the sameperiod and included all eligible women with arecorded post partum blood loss of at least600 ml after a normal, unassisted vaginal deliv-ery. Controls were matched to cases by thefacility where the case originally was booked todeliver and by the week of the delivery. Casedata were abstracted from the woman's medi-

Table 1 Risk factors and their relative (RR) estimates from the logistic regressionmodels

Risk factor Estimate of 95%the adjusted * confidenceRR interval

Cephalopelvic disproportion:Age 35 + 21 (0-96, 4 4)Height < 160 cm 2-0 (1-3, 3-0)Nulliparous 13 8 (7-3, 25 9)Operative delivery in last pregnancy 9 5 (3-8, 23 6)Neonatal death in last pregnancy 4 5 (1 3, 15 9)Antenatal hospitalisation for PIH 6.8 (1 6, 28 3)

Post partum haemorrhage:Age 35+ 2 6 (1.2, 5-8)Low parity (0, 1) 1 7 (1 1, 2 7)Poor obstetric outcome last pregnancyt 1 8 (0 94, 3 7)Antenatal haemoglobin < 12 g/dl 2 2 (0-99, 5 0)Antenatal hospitalisation forpregnancy related problem 4-3 (1 4, 12 8)

* Adjusted for the other factors listed and for facility booked (on which cases and controls were

matched).t Post partum haemorrhage, 1st or 2nd trimester, fetal death, stillbirth, neonatal death.PIH = pregnancy induced hypertension.

cal record while she was still in the hospital,while data for controls were abstracted withina few weeks after delivery from the uniformmaternal record form used for all deliveries inthe public system.To evaluate the effectiveness of various fac-

tors in screening pregnant women for hospitalrather than home delivery, only factors detect-able before the onset of labour were used. Riskfactors identified for cephalopelvic dispropor-tion included advanced maternal age (35 yearsand older), short stature (less than 160 cm),nulliparity, a history of operative delivery orneonatal death in the preceding pregnancy,and admission to hospital for pregnancyinduced hypertension in the current preg-nancy. Those identified for post partum haem-orrhage also included advanced maternal ageand low parity, but included a more generalhistory of poor obstetric outcome in the pre-ceding pregnancy, anaemia during the currentpregnancy, and admission to hospital for anypregnancy related problem before the onset oflabour as well. Table 1 lists the factors andtheir adjusted relative risk (RR) estimates fromthe two logistical models.

Several approaches were undertaken to de-vise screening tests using these factors and toevaluate the resulting systems. Initially, thelogistic models were used to calculate esti-mated probabilities for each subject for the twoseparate outcomes (see Appendix for an expla-nation of the computations). In addition, avariety of different algorithms were developedusing cumulative scores based on the sum of allrisk factors, either unweighted or weighted bytheir RR values, or simple tests based on thepresence of one or more selected risk factors,alone or in combination with each other.Screening tests were devised for each of thetwo complications separately and, since a goodscreening system should identify women atrisk for either of these problems, for a pooledpopulation consisting of both case groups andthe controls. Factors used for screening thepooled population were derived from thoseelements the two case studies had in common.

Sensitivity, specificity, and likelihood ratios(the ratio of sensitivity to 1-specificity) werecalculated for the dichotomous algorithms andfor various cut off points for the continuousscores and the probabilities. All logical com-binations of factors were tested, but only thosewith higher likelihood ratios were reportedhere. Receiver operating characteristic (ROC)curves, plots of the true positive rate (sensiti-vity) against the false positive rate (1-specifi-city), were derived for the probabilities esti-mated from the logistic models and for theweighted and unweighted cumulative scores.

Predictive values for a test could not bederived from this case-control study directly,but were calculated using the estimated inci-dences of cephalopelvic disproportion and postpartum haemorrhage in this population. Onlylikelihood ratios and predictive values for pos-itive tests are considered here, since negativetests in this particular setting will result in nochange in the course of action planned beforethe test (that is, either home or hospital deliv-

Tsu



j.com/

J Epidem

iol Com

munity H


nloaded from


How useful are obstetric risk factors?

ery). The proportion of the population thatwould be referred according to a particular testconstitutes an important measure of the "cost"of the test to the system. Like the predictivevalues, it was calculated using the estimatedincidences of cephalopelvic disproportion andpost partum haemorrhage in this population.

ResultsScores for cases and controls, based on sums ofthe risk factors, either unweighted or weightedby the RR values, were compared using ROCcurves. For cephalopelvic disproportion (seefig 1), the score based on weighted factorsperformed modestly better than one usingunweighted factors, but even at its optimumpoint (the point on the curve closest to theupper left corner), that is using a cut off of 4 ormore, the weighted score had both a sensitivityand specificity of only 75%. The curves forpost partum haemorrhage (fig 2) were only abit above the diagonal line representing resultsthat could be expected by chance if the testwere simply random, with scores based onweighted factors showing little advantage overunweighted ones.The cumulative scores and dichotomous

100A Unweighted factors

80-

.5 60-

60 Chance line(D 40 -2

(I)

algorithms were compared on the basis ofsensitivity, specificity, likelihood ratio (LR),positive predictive value (PV+), and "cost"(total proportion of the population referred).Table 2 shows the comparative utility of sys-tems for cephalopelvic disproportion and postpartum haemorrhage separately, in ascendingorder of cost. Because of the low incidences ofthese two conditions in this population, thepositive predictive values are quite low, withnone above 7%. Focussing on nulliparouswomen who are less than 160 cm tall, however,would result in referrals of only 4% of allpregnant women and would identify 28% ofthe cephalopelvic disproportion cases. Usingthe presence of any two of the significant riskfactors as the screening criteria achieves bettersensitivity without too great a reduction inspecificity and would result in just under 10%of women being referred and more than 40%of cephalopelvic disproportion cases beingidentified. Combining the strongest predic-tors, nulliparity (conditional on short stature)and operative delivery in the last pregnancy,yields nearly identical results in this popula-tion. At the high cost end, referring onequarter of the pregnant women (using thescore with weighted factors and a cut off point

U)(I)

1-Specificity

U)

(I,

.1

1-Specificity

Figure 1 Receiver operating characteristic (ROC)curves for cephalopelvic disproportion based on

cumulative scores using sums of unweighted risk factorsand sums of risk factors weighted by the relative risk(RR) values, at selected cut off points.

C,,t_lU)(a)n)

0 20 40 60

1-Specificity

80 100

0 20 40 60 80 100

1-Specificity

Figure 2 Receiver operating characteristic (ROC)curves for post partum haemorrhage based on cumulativescores using sums of unweighted risk factors and sums ofrisk factors weighted by the relative risk (RR) values,at selected cut off points.

299



j.com/

J Epidem

iol Com

munity H


nloaded from


300

of 4 or more) would result in three quarters ofthe cephalopelvic disproportion cases beingpredicted.As was seen with the ROC curves, the post

partum haemorrhage predictors are not asstrong as those for cephalopelvic dispropor-tion. The highest LR is achieved when thecriteria consist of either poor obstetric out-come in the last pregnancy or antenatal admis-sion to hospital for a pregnancy related prob-lem. This combination identifies nearly a thirdof the cases, at a cost of only 5 6% of womenbeing referred. Adding a low haemoglobinconcentration to this combination increases

Table 2 Utility of various predictive criteria for cephalopelvic disproportion and postpartum haemorrhage separately

Score system % Specificity Likelihood Predictive % of(incidence of predicted ratio value of populationcondition) (sensitivity) positive referred

test (%) ("cost")

(A) Cephalopelvic disproportion ( 010):(1) Nulliparous + height

<160 281 960 70 66 42(2) Any 2 or more risk

factors 42 3 90 5 4-5 4 3 9 8(3) (Nulliparous + height

<160) or CPD Hx* 40 9 900 41 40 10 3(4) #3 or NND Hxt or

PPH or (age 35 + andheight < 160) 49 8 86 3 3 6 3 5 14 1

(5) Nulliparous 53 7 85 3 3 7 3 6 15 1(6) Factors weighted by RR

(cut= 13+) 562 847 37 36 157(7) '6 (cut=4+) 722 752 30 29 253

(B) Post partum haemorrhage ( 008):(1) Factors weighted by RR

(cut=4+) 127 970 42 33 3 1(2) ObHx I * or antenatal

hospitalisation 30-0 94 6 5 6 4 3 5 6(3) Any 2 or more risk

factors 15 4 92-5 2 1 1-7 7 6(4) ObHxlt or antenatal

hospitalisation or lowhaemoglobin 35-0 90 3 3-6 2 8 9 9

(5) #1 (cut=2+) 315 824 18 14 177(6) Any 1 or more risk

factor 76 5 52 2 1-6 1-3 48 0

* CPD Hx: operative delivery in last pregnancy t NND Hx: neonatal death in last pregnancy+ ObHxl: post partum haemorrhage, 1st or 2nd trimester fetal death, stillbirth, or neonatal deathin last pregnancy.

Table 3 Utility of various predictive criteria for pooled outcomes (cephalopelvicdisproportion and post partum haemorrhage)

Score system % Specificity Likelihood Predictive % of(incidence predicted ratio value of populationof condition) (sensitivity) positive test referred

(%) ("cost")

(A) Pooled ( 018):(1) (Nulliparous + height < 160)

or ObHx2* or PIH 34-5 87-6 2 8 4.9 12 8(2) Nulliparous 39-0 85 3 2-7 4-7 15 1(3) Nulliparous + height

< 160)or ObHx2* or antenatalhospitalisation or(age 35 + or lowhaemoglobin) 42 4 82 3 2-4 4 2 18 1

(B) Nulliparous ( 046):( 1) Antenatal

hospitalisation 52-2 65-2 1 5 6 7 35 0

(C) Muitiparas ( 013):(1) ObHx2* 22 5 910 2-5 32 101(2) ObHx2* or antenatal

hospitalisation 35.3 85-3 2 4 3-1 15 3

* ObHx2: operative delivery, neonatal death, or post partum haemorrhage in last pregnancy.

the sensitivity to 35% but nearly doubles thecost to 10% of women being referred. Al-though the weighted score has a likelihoodratio of 4 2 at a cut off value of 4 or more, thesensitivity is so low as to be useless, and at alower cut off of 2+ (LR= 1 8) it performsmuch worse than the simple combination ofhistory and/or antenatal hospital admission.To identify three quarters of the post partumhaemorrhage cases with these factors wouldrequire the referral of nearly half of all preg-nant women.Table 3 compares the predictive utility of

these factors with regard to the two outcomesin a pooled population, as well as for pooledoutcomes in two parity subgroups (nulliparasand multiparas). As one might expect, usingthe score systems for a pooled set of complica-tions results in lower specificity and, thus,lower LRs. Nulliparity alone performed aboutas well as or better than any of the other morecomplicated algorithms, but it would haveidentified less than 40% of the cases compli-cated by either cephalopelvic disproportion orpost partum haemorrhage. A combination of(1) shorter, nulliparous women, (2) those withan operative delivery, post partum haemor-rhage, or neonatal death in the last pregnancy,and (3) those admitted to hospital for preg-nancy induced hypertension during the cur-rent pregnancy yields a slightly higher LRthan nulliparity alone, but results in a drop insensitivity in exchange for its slightly reducedcost. Sensitivity can be increased to 42 4% bybroadening the indications for antenatal hos-pital admission and adding older women with alow haemoglobin concentration, but specifi-city drops then and the cost climbs sharply.Even with all these limitations, in a pooledpopulation the cost can be kept at about 13%of pregnant women being referred while pre-dicting nearly 35% of the complicateddeliveries.The different incidence rates among the

parity subgroups have a modest impact on thepositive predictive values. For example, al-though antenatal hospital admission is neithervery sensitive nor very specific among nulli-parous women (evident in its low LR of 1-5),its positive predictive value is the highest ofany of the score systems, primarily because theincidence of complications is so much higherin this population group. Despite the strengthof obstetric history as a risk factor, its lowsensitivity and the low rate of cephalopelvicdisproportion and post partum haemorrhageamong parious women greatly limit the posit-ive predictive value of scoring systems in thisgroup.When individual probabilities for the com-

plications of interest were estimated from thelogistic models, actual cases did generally havehigher probability scores than controls, butnone of the probabilities was very high andthere was considerable overlap between valuesfor cases and controls (especially for post par-tum haemorrhage). For example, 80 6% ofpost partum haemorrhage cases and 92 2% ofcontrols had estimated probabilities for postpartum haemorrhage less than 0 02. Since

Tsu



j.com/

J Epidem

iol Com

munity H


nloaded from



these probabilities are computationally morecomplex to generate and offer little advantageover simpler screening algorithms, no furtherevaluation of them is provided in this report.

DiscussionThe value of these tests for antenatal screeningclearly depends on the criteria by which theyare judged. There is no single measure thatsatisfies all concerns. One of the main prob-lems is that almost all measures are influencedby the characteristics of the population towhich they are applied. Another is that botheffectiveness and cost (in a very broad sense)must be taken into account.

MEASURES OF UTILITYSensitivity and specificity are basic measuresof effectiveness that can characterise a testindependently of the incidence of the targetcondition, but they can be affected by varia-tions in the frequencies of the particular riskfactors used. Attempts to combine sensitivityand specificity into one measure (called the"accuracy"9 or the "percent correctly classi-fied"5) can be quite misleading'0 and thus werenot used here. The trade off between sensiti-vity and specificity was particularly evident inthis study, since the sensitivity of individualfactors was low, but more effective com-binations of factors were inevitably lessspecific.The low predictive values for all the screen-

ing tests in the present study are to beexpected, at least to some extent, because ofthe low incidence of cephalopelvic dispropor-tion and post partum haemorrhage in Harare.With an incidence of2% or less, even a screen-ing test with 95% sensitivity and 95% specifi-city could not achieve a positive predictivevalue better than 28%. With regard to cepha-lopelvic disproportion, there is reason tobelieve (Munjanja, personal communication,1991) that there might have been a somewhathigher incidence in Harare if women with twoor more prior caesarean sections or other riskfactors for cephalopelvic disproportion werenot given elective caesareans, but even thatadjustment would not push the incidenceabove 2%. It is nearly impossible to get popu-lation based estimates of the incidence ofcephalopelvic disproportion in Africa, but thefigure for Harare is consistent with availabledata from other studies.'"'5 While some hos-pital based studies have reported rates as highas 15%,16 hospital populations usually overrepresent women with problems, a bias that isparticularly true among "unbooked" patients(those not seen for antenatal care, who come tothe hospital only when a complication occurs)and may also be true of "booked" patients,since these women may be self selectedbecause of prior or anticipated delivery com-plications. It seems unlikely that the trueunderlying rate of cephalopelvic disproportionis much more than 2% in most places, so onecannot expect substantial improvement in pre-dictive values for cephalopelvic disproportionin other populations.

There is some basis, though, for thinkingthat the incidence of post partum haemorrhagein Harare is considerably lower than the usualrange of rates (10-20%) reported elsewhere.'7This is partly a result of the exclusion ofoperative deliveries in this study and perhapsof the routine administration of syntometrine(a combination of oxytocin and ergometrine)after delivery to ensure prompt contraction ofthe uterus. Prendiville, Elbourne, andChalmers,'8 based on a review of data fromcontrolled clinical trials, have estimated thatactive management of the third stage of labour(including the use of oxytocics) reduces theincidence of post partum haemorrhage byabout 40%. Begley'9 compared post partumhaemorrhage rates with and without oxytocicsin a controlled trial in the UK and found a fourfold higher rate without oxytocics (8% versus2%). Such a ratio suggests a higher underlyingrate of post partum haemorrhage in popula-tions without intervention, very similar to the8-4% rate found in a study of births attendedby traditional birth attendants in Malawi."0The rate of post partum haemorrhage in theNigerian hospital series2' was 2-8%. If themore conservative correction factor of 40%held true in Harare, the incidence of postpartum haemorrhage one could expect if oxy-tocics were not in use would be 1-3% instead of0-8%. If higher underlying rates of 8-10%were typical, some improvement in predictivevalues could be expected.The lack of difference in predictive values

between scores among nulliparous and parouswomen, despite the higher likelihood ratios forscores using obstetric history, is certainlypartly due to the lower incidence of cephalo-pelvic disproportion and post partum haemor-rhage in parous women. Alexander andKierse22 point out that although most studiesfind scoring to be more predictive in multipar-ous women, several exceptions to that havebeen reported previously. Also, scoring sys-tems that rely more on socioeconomic factorsgenerally find little difference in predictivevalue between nulliparous and parous women.

Cost considerations include the medical,financial, and social burdens associated withmissed cases, unnecessary interventions, andadministration of the test itself, as those bur-dens apply to individual women and theirfamilies and to the health care system. Thesecosts are nearly impossible to quantify, butthey can at least be compared on a relativebasis to the current practice or to other screen-ing options being considered. A major limi-tation on the usefulness of likelihood ratios andpredictive values when making managementdecisions is that they give equal weight to falsepositives and false negatives, an approach thatseldom reflects reality. In the present study,the costs of a missed case are high to thewoman and the system, because both cephalo-pelvic disproportion and post partum haemor-rhage are life threatening complications andare more expensive and difficult to manage ifnot treated promptly. The costs of unneces-sary intervention, where intervention consistsof referral for delivery in a facility where

301



j.com/

J Epidem

iol Com

munity H


nloaded from


302

operative delivery and medical or surgicalmanagement of post partum haemorrhage areavailable, are also high in most parts of thedeveloping world, where there are not enoughpersonnel or facilities to accommodate all de-liveries and many women cannot easily reachor afford care in an appropriate facility whenthe time for delivery arrives. Where healthsystem capabilities are limited, unnecessaryreferrals may either overwhelm the system orpre-empt space needed for women who cantruly benefit from hospital care. As was done inan earlier study in Zaire,23 this study definedcost as the percentage of women referred (thatis, test positives) in the population. The costsof the tests themselves are negligible, given thetype of screening systems discussed in thisstudy (since they involve data already collectedas part of routine physical examination andmedical history taking), as long as antenatalcare is already provided for purposes otherthan screening. Aside from haemoglobin de-termination, the factors are observational andnot invasive.Where the capacity of the health system is

the limiting factor, screening tests must beevaluated with that consideration foremost. Itis useful to choose a particular level of referralsthat can be tolerated (Mdller24 suggested10-20% of women, for example) and thencompare different tests on the basis of theproportion of cases they can detect within thatreferral limit. If referrals for hospital deliverywere capped at 10% of pregnant women, ascreening test that was considered positive if awomen had two or more of the cephalopelvicdisproportion risk factors present (test (A)2 ontable 2) would be the best and would detect42 3% of cases. A screening test based on thepresence of poor obstetric history, antenatalhospital admission for a pregnancy relatedproblem, or haemoglobin less than 12 g/dl (test(B)4 on table 2) would be the best test for postpartum haemorrhage within the 10% cap, butwould detect only 35% of cases. None of thetests for a pooled outcome of cephalopelvicdisproportion and post partum haemorrhagecould meet the ceiling of 10% referred, al-though the screening test that combines short,nulliparous women, women with poor ob-stetric history, and women with pregnancyinduced hypertension ((A)1 on table 3) doesnot exceed it by much and identifies about35% of cephalopelvic disproportion and postpartum haemorrhage cases.

COMPARISON WITH OTHER STUDIES OFOBSTETRIC SCREENINGWhile several studies in Africa have examinedpossible risk factors for obstetric complica-tions, only a handful have looked systemati-cally at the predictive values of the risk factorsfor maternal outcome. Two studies focussedon height and cephalopelvic disproportion. InTanzania, primigravid women with heights of146 cm or less accounted for 32% of the popu-lation and 90% of the caesarean sections forcephalopelvic disproportion (LR = 3 0). If thecut off were set at 141 cm or less, only 11% of

the population would be included while stillidentifying 67% of the cephalopelvic dispro-portion cases." Using a cut off of 60 inches orless (150 cm) for primigravidae in SierraLeone, Aitken and Walls25 had to refer 57% ofpregnant women in order to identify 84-6% ofcephalopelvic disproportion cases (LR = 0 5).Using the data from the present study and acut off of 160 cm in nulliparous women, thepredictive efficacy is midway between the Tan-zania and Sierra Leone examples (52-3% ofcases detected for 14 9% of the populationreferred, LR= 1 9).Some studies have combined height with

other factors. In Tanzania, a test with 18possible factors26 identified 23 2% of pregnantwomen as "at risk" while detecting 54 3% ofthe cases of maternal complications at delivery(LR = 26). Also in Tanzania, pelvic assess-ment was used in primigravidae27 to detect584% of cephalopelvic disproportion caseswhile referring just 21 8% for hospital delivery(LR = 42). In Zaire, two studies have linkedrisk factors with outcome. The Kasongo Pro-ject Team23 compared risk factors recordedantenatally with two outcomes: life threateningfetopelvic disproportion and abnormally pro-longed labour. They found that a history ofprevious dystocia or of previous perinataldeaths was significantly associated with bothoutcomes. They tested three approaches torisk prediction: (A) single factor, (B) any ofseveral factors, and (C) specific combinationsof factors. Among the single factors, a badobstetric history was the most effective (sensit-ivity of 18%) at the lowest cost (3% referred).This is much better than a similar historyfactor in the current study, which identifiedonly 16% of cephalopelvic disproportion caseswith a cost of 9% referred. With model (B),using positivity for any one of several factorsincluding a bad obstetric history, fetal posi-tion, and maternal age, height, and gravidity,the effectiveness rose to 69% of cases detectedbut cost rose to 44% of the populationreferred. With model (C), if either a bad obste-tric history or young age with short heightwere present, effectiveness was 59% with acost of 18% referred. In Karawa, Zaire,screening tests to identify those needing surgi-cal delivery because of obstructed labour werebased on parity, height, and obstetric history,but the most efficient test was based on heightless than 145 cm.'6 By referring 13% of preg-nant women, 46% of women with cephalopel-vic disproportion could be identified. This isroughly comparable with the results achievedwith test (A)2 on table 2 (the presence of anytwo or more risk factors). At the high end ofcost, if 26% of women were referred (forheight < 150 cm), 63-7% of cephalopelvic dis-proportion cases were identified in Karawa.The use of height alone results in a slightlyhigher yield than when other factors are in-cluded, as in test (A)7 on table 2. This test,which consists of a score summing all riskfactors, weighted by their RR values and usinga cut off point of 4 or more, identifies nearly18% more cases than the Karawa test (75%versus 64%) while referring the same propor-

Tsu



j.com/

J Epidem

iol Com

munity H


nloaded from



tion of pregnant women. The predictive valuesin the Karawa study are much higher than inthe present study, though, because the inci-dence of cephalopelvic disproportion was somuch higher in their hospital based sample.The Karawa study seems to be the only onepublished to date in which estimated probabi-lities (based on individual combinations of riskfactors) using multiple logistic regression ana-lysis have been calculated. Their derivationfrom a hospital based sample limits their appli-cation, but they do show a consistent trend inwhich the proportion of women with cephalo-pelvic disproportion is higher in groups withhigher estimated probabilities.

Elsewhere in the developing world therehave been some efforts to use risk factors toscreen pregnant women. In Papua New Gui-nea, Lennox28 found 53% of women with bothhospital and home based deliveries had at leastone high risk factor and accounted for 70% ofthe complicated deliveries. A history of pre-vious third stage of labour complications alonepredicted 17% of subsequent third stage com-plications at a cost of just 3 5% of womenreferred. In a study in the Philippines,Esguerra et aP9 applied a score with antenataland intrapartum components to a group ofwomen and measured maternal and perinatalmorbidity: while sensitivity was high, the spe-cificity was low and the likelihood ratios werenot much above 1 0. A Thai study30 tested a 43item- score against "abnormal deliveries" andachieved a sensitivity of 28-8%, specificity of84-4%, and LR of 1-8. Studies in other Africansettings31-35, in Australia,36 South America,37 38and Asia3'41 describe risk screening pro-grammes but fail to link risk status with actualoutcome.While risk systems in the developing world

have dealt with both maternal and perinataloutcomes, those in the developed world dealalmost exclusively with perinatal outcomes.(perhaps because poor maternal outcome isrelatively rare). Several reviews of these sys-tems have been published that point out theweaknesses and difficulties of the researchdone so far, including the absence of basicepidemiological statistics (such as sensitivity,specificity, or incidence rates) in many studies;the failure to acknowledge the potential impactof characteristics such as the age and ethnicdistribution of particular study populationswhich limits their use in other groups; and thedependence of many systems on intrapartuminformation such as gestational age and birth-weight (factors that are of little value in guid-ing management since they are detected toolate for effective intervention).4245The ultimate value of any risk screening

system depends not only on its effectivenessand cost but on the effectiveness of availableinterventions and the ability of the healthsystem to implement them. In the cases ofcephalopelvic disproportion and post partumhaemorrhage, appropriate interventions (sur-gical and medical management) are well knownand, at least in Zimbabwe, are available at mostdistrict hospitals. In places where such ser-vices are not available, any attempts to insti-

tute risk screening programmes must be ac-companied by plans to make the requisiteinterventions accessible to the population. Forother obstetric complications, such as preg-nancy induced hypertension or pretermlabour, for which the interventions are morecomplex or less effective, the value of riskscreening would be reduced.

STUDY LIMITATIONS AND STRENGTHSThere are several limitations inherent in thisstudy and its setting that constrain the conclu-sions one can draw. One of the most significantis that the risk models were tested only on thedata set from which they were drawn (as is trueof the vast majority of such studies), a processthat inflates their predictive values.22 An im-portant follow up to this study would be toapply the screening tests to other data sets andsee how well they do. Another limitation is thefact that the study is based on an urban popu-lation, even though the most urgent need forsuch risk screening is in the rural areas whereambulances and local hospitals are not avail-able. This urban context may limit the exten-sion of its findings to rural populations inAfrica or other areas of the developing world.Although these women all had trained medicalpersonnel attending their deliveries, a fact thatdistinguishes them from the vast majority ofrural women, these deliveries involved a verylow level of medical intervention, quite similarto the care rural women might receive. Thereis no obvious reason to believe that the riskfactors identified in this study would differ forurban and rural residents. The common shift-ing of residence between urban and rural areasin Zimbabwe makes it less likely that Hararewomen differ substantially from their ruralcounterparts. If there were differences in thedistribution of risk factors such as age orparity, they could affect the predictive valuesof the screening tests proposed. In particular,in a population where elective caesarean sec-tion because of a history of two or moreprevious sections is not common (unlike Har-are, where it is official policy and nearly uni-versally applied), the importance of nulliparityas a risk factor for cephalopelvic disproportionmight drop considerably while the importanceof a history of previous caesarean sectionmight rise correspondingly. As a result,screening tests using nulliparity might be lesseffective while those using history might per-form better.A further limitation was the narrow focus of

the study on two particular complications-cephalopelvic disproportion and post partumhaemorrhage. Any useful antenatal screeningsystem would have to address all possibleserious obstetrical complications known to oc-cur in significant numbers. In attempting to doso, the specificity of the test might declinesomewhat (although this might be offset interms of predictive value by a higher combinedincidence rate). The more narrow systems de-scribed here may represent an overestimate ofwhat can be achieved with risk screening. Onthe other hand, if the risk factors for cephalo-

303



j.com/

J Epidem

iol Com

munity H


nloaded from


304

pelvic disproportion and post partum haemor-rhage are also related to other obstetric andperinatal complications, the yield in terms ofpredictive value could be much higher in acombined screening test than in the onesshown here.The conclusions to be drawn from this study

regarding the value of risk screening arestrengthened by several factors. Firstly, thescreening tests were applied to groups repres-entative of the general population of pregnantwomen in Harare, not just to a higher riskhospital population as in most other studies inAfrica. The outcomes (cephalopelvic dispro-portion and post partum haemorrhage) werecarefully defined and systematically ascer-tained from a uniform municipal medical re-cord. The two outcomes were not subject toalteration as a result of early risk status identi-fication, since they could only be treated andnot prevented (except possibly for thosewomen with two or more previous caesareansections assigned to elective section for sus-pected cephalopelvic disproportion).The results of this study suggest that a third

or more of the women likely to experiencecephalopelvic disproportion or post partumhaemorrhage can be identified sufficiently inadvance of the onset of labour to enable themto plan to deliver at a facility equipped tohandle such problems. This can be done at arelatively modest cost, in terms of the testingitself and the excess of unnecessary referrals tothe hospital, as long as antenatal care is alreadyavailable or such screening can be incorpor-ated into another community health care struc-ture (such as traditional birth attendants orcommunity health workers). Further refine-ments to increase the sensitivity or specificityof the screening tests or to determine riskfactors that are common to other adverse out-comes (thereby increasing the pertinent inci-dence rates) would enhance the effectiveness ofthe tests. The continued expansion and im-provement of the primary health care struc-ture, including surgical services and bloodtransfusion capabilities relevant to other healthproblems besides obstetrics, would reduce theunit costs, both of administering the tests andof dealing with excess referrals, by broadeningthe service base and spreading it over severalhealth functions. The screening models pre-sented here provide a basis for comparing thevalue of screening for these conditions (cepha-lopelvic disproportion and post partum haem-orrhage) with the results to be expected fromalternative strategies, such as screening forother conditions of concern, treating sympto-matic women only, or developing emergencytransport or care capabilities in everycommunity.

AppendixThe conditional probability (p) for an event

(either cephalopelvic disproportion or postpartum haemorrhage), given a set of values forthe selected risk factors, was based on the factthat

p = el°git P (1 + elogi't P)

and for a case control study

logit p = ln (p[sampled case] p[sampled control])+a+IXI+ ...+ kXk

where alpha (a) represents the case-controlintercept value, from which a sampling correc-tion term - -ln (p[sampled cCase] p[sampled ]) mustbe added to get a cohort/based intercept, andeach beta (p) and x represent a set of coeffi-cients and factors. The numerator of the sam-pling correction term is 1, since 100% of thecases were included. The proportion of con-trols sampled varied somewhat by facility ofbooking (23% of clinic, 0 6% of hospital, and0-8% of unbooked). The logit was, therefore,adjusted by a different weighting factordepending on a given control's booking status.

I would like to express appreciation for the Fulbright Fellow-ship which supported the data collection in Zimbabwe andspecial thanks to Dr Noel S Weiss for his critical advice andconstant encouragement throughout the research and prepara-tion of this manuscript.

1 Adelstein P, Fedrick J. Antenatal identification ofwomen atincreased risk of being delivered of a low birth weightinfant at term. Br J Obstet Gynaecol 1978;85:8-1 1.

2 Aubry RH, Nesbitt REL. High-risk obstetrics. AmJ7 ObstetGynecl 1969;105:241-7.

3 Goodwin JW, Dunne JT, Thomas BW. Antepartum identi-fication of the fetus at risk. Can Med Assoc J71969;101 :458-64.

4 Hobel CJ. Recognition of the high-risk pregnant woman.In: Spellacy WN, ed. Management of the high-risk preg-nancy. Baltimore: University Park Press, 1976; 1-28.

5 Kaminski M, Goujard J, Rumeau-Rouquette C. Predictionof low birthweight and prematurity by a multiple regres-sion analysis with maternal characteristics known since thebeginning of pregnancy. Int Jf Epidemiol 1973;2:195-204.

6 Royston E, Lopez AD. On the assessment of maternalmortality. World Health Stat Q 1987;40:214-24.

7 Tsu VD. Maternal height and age: Risk factors for cephalo-pelvic disproportion in Zimbabwe. Int J Epidemiol1992;21:941-6.

8 Tsu VD. Postpartum haemorrhage in Zimbabwe: A riskfactor analysis. Br J Obstet Gynaecol 1993;?:

9 Grant A, Mohide P. Screening and diagnostic tests inantenatal care. In: Enkin M, Chalmers I, eds. Effectivenessand satisfaction in antenatal care. London: William Heine-man Medical, 1984; 22-59.

10 Weiss NS. Clinical epidemiology: The study of the outcome

of illness. Monographs in epidemiology and biostatistics. Vol1. Oxford: Oxford University Press, 1986; 21.

11 Everett VJ. The relationship between maternal height andcephalopelvic disproportion in Dar es Salaam. East Af-rican MedJ7 1975;52:251-6.

12 Fuchs K, Peretz, B-A, Marcovici R, Paldi E, Timor-TritshI. The 'grand multipara'. Is it a problem? A review of 5785cases. Int J Gynecol Obstet 1985;23:321-5.

13 Howard D. Aspects of maternal morbidity: the experienceof less developed countries. In: Jelliffe D, Jelliffe P, eds.Advances in international maternal and child health.Oxford: Clarendon, 1987; 1-35.

14 Liljestrand J, Bergstrom S, Westman S. Maternal heightand perinatal outcome in Mozambique. 7 Trop Pediatr1985;31 :306-10.

15 Mtimavalye LAR, van der Does CD, Maathuis JB. Therelationship between increasing birthweight and cephalo-pelvic disproportion in Dar es Salaam, Tanzania. 7 ObstetGynaecol Br Commonwealth 1974;81:380-2.

16 Hermann CB, Duale S. Improving maternity care services in

the Karawa Health Zone of Zaire. Maternal Nutrition andHealth Care Program, Report No 2. International Centerfor Research on Women: Washington DC, 1990.

17 World Health Organization (WHO). The prevention andmanagement of post partum haemorrhage. Report of a Tech-nical Working Group, WHO/MCH/90.7, Geneva: WHO,1990.

18 Prendiville W, Elbourne D, Chalmers I. The effects ofroutine oxytocic administration in the management of thethird stage of labour: An overview of the evidence fromcontrolled trials. Br_7 Obstet Gynaecol 1988;95:3-16.

19 Begley CM. A comparison of 'active' and 'physiological'management of the third stage of labour. Midwifery1990;6:3-17.

20 Bullough CHW, Msuku RS, Karonde L. Early sucklingand post partum haemorrhage: Controlled trial in deliver-ies by traditional birth attendants. Lancet 1989;2:(8662):522-5.

21 Harrison KA. Background information. In: Harrison KAed. Child-bearing, health and social priorities: a survey of22,774 consecutive hospital births in Zaria, NorthernNigeria. Br J Obstet Gynaecol 1985;5(suppl):3-13.

22 Alexander S, Keirse MJNC. Formal risk scoring during

Tsu



j.com/

J Epidem

iol Com

munity H


nloaded from



pregnancy. In: Chalmers I, Enkin M, Kerise MJNC, eds.Effective care in pregnancy and childbirth. Vol 1. Oxford:Oxford University Press, 1989; 345-65.

23 Kasongo Project Team. Antenatal screening for fetopelvicdystocias: a cost-effectiveness approach to the choice ofsimple indicators for use by auxiliary personnel. J7 TropMed Hyg 1984;87(4):173-83.

24 Moller B. The outcome of pregnancy and antenatal care inrural Tanzania. Acta Universitatis Upsaliensis, Comprehen-sive summaries of Uppsala dissertations from the Faculty ofMedicine, 177. Uppsala: Uppsala University, 1985.

25 Aitken IW and Walls B. Maternal height and cephalopelvicdisproportion in Sierra Leone. Trop Doct 1986;16:132-4.

26 Walther FJ. At-risk selection and outcome studies at theKibara Hospital antenatal clinic in Tanzania. Trop GeogrMed 1980;32(4):336-9.

27 Bauer 0, Kingu R, Laussen T, Mbwana K. Pelvic assess-ment and cephalopelvic disproportion in Central Tanza-nia. Int J Gynecol Obstet 1988;27:323-5.

28 Lennox CE. Assessment of obstetric high risk factors in adeveloping country. Trop Doct 1984;14:125-9.

29 Esguerra AB, Diamante AN, Rams MM, et al. Concisescreening scales for high-risk mothers and newborns. In:Del Mundo F, Ines-Cuyegkung E, Aviado DM, eds.Primary maternal and neonatal health: A global concern.New York: Plenum, 1983; 347-69.

30 Sirivongs B, Parisunyakul S. Risk pregnancy screening: Asimple method for non physicians to screen the high-riskpregnancy. Jf Med Assoc Thai 1984;67(suppl) 2):15-20.

31 Essex BJ, Everett VJ. Use of an action-orientated recordcard for antenatal screening. Trop Doct 1977;7:134-8.

32 Sogbanmu MO. Perinatal mortality and maternal mortalityin General Hospital, Ondo, Nigeria: Use of high-riskpregnancy predictive scoring index. Nigerian Med Jf1979;9:475-9.

33 Malone MI. The quality of care in an antenatal clinic inKenya. East African MedJ7 1980;57:86-89.

34 Kennedy I. Botswana: Midwives use antenatal cards todetect at-risk pregnancies. Salubritas 1982;6(4):1,7.

35 Leke RJ, Nasah BT, Mtango FDE. Introduction of highrisk pregnancy care in rural Cameroon: Health serviceresearch approach. Journal of Obstetrics and GynaecologyEast Central Africa 1988;7:7-10.

36 Watson DS. Use of "at risk" antenatal score card byaboriginal health workers in Arnhem Land. Trop Doct1984;14: 133-5.

37 Araujo G, Araujo L, Janowitz B, Wallace S, Potts M.Improving obstetric care in Northeast Brazil. Pan Amer-ican Health Organisation Bulletin 1983;17:233-42.

38 Kirschbaum A, Salomon A, Parker E et al. Pregnancy riskfactor assessment-North area of Santiago, Chile, 1982-1983. MMWR 1985;34:335-7.

39 Shah KP. Surveillance card for married women for betterobstetric performance. Journal of Obstetrics and Gynae-cology India 1978;28:1015-20.

40 Kumar V, Datta N. Home-based mothers' health records.World Health Forum 1988;9:107-10.

41 Alisjahbana A, Peeters R, Meheus A. Traditional birthattendants can identify mothers and infants at risk. WorldHealth Forum 1986;7:240-2.

42 Herman AAB, Irwig LM, Groeneveld HT. Evaluatingobstetric risk scores by receiver operating characteristicscurves. Am J Epidemiol 1988;127:831-42.

43 Marshall VA. A comparison of two obstetric risk assessmenttools. J Nurse Midwifery 1989;34:3-7.

44 Morrison I, Carter L, McNamara S, Cheang M. A simpli-fied intrapartum numerical scoring system: The predictionof high risk in labor. Am Jf Obstet Gynecol1980;138:175-80.

45 Selwyn BJ. Review of obstetrical risk assessment methods.In: Research issues in the assessment of birth settings. Wash-ington, DC: National Academy Press, 1982; 149-70.

305



j.com/

J Epidem

iol Com

munity H


nloaded from


antenatal screening: its in assessing obstetric risk ...journalofepidemiology andcommunityhealth...

Documents