Table 1

Studies estimating the validity or reliability of women's self-report of healthcare personnel assisting during delivery

Reference	Study setting, participants and sample size	Recall period	Source of comparison against women’s self-report	Statistical measures	Key findings
Blanc et al. 2016a21	Public hospital in Mexico City. Pregnant women aged 15–49 years admitted to the study facility for delivery. n=597	Interviews conducted with women prior to hospital discharge.	Direct observation by general medical practitioners or nurses.	For individual reporting accuracy: sensitivity, specificity and AUC* For population-based validity: IF†	Main provider at delivery was SBA (doctor or medical resident) had a high sensitivity (90.1%) and a low specificity (14.0%), resulting in low individual-level accuracy (AUC: 0.52, 95% CI: 0.48 to 0.56) and low population-level bias (IF: 0.98). Vast majority of participants reported the main provider during delivery was a doctor or a medical resident (94%); both cadres are considered SBA.
Blanc et al. 2016b18	Two public hospitals in Kisumu and Kiambu districts, Kenya. Pregnant women aged 15–44 years admitted to study facilities for labour and delivery. n=662	Interviews conducted with women prior to hospital discharge.	Direct observation by registered nurse/midwives.	For individual reporting accuracy: sensitivity, specificity and AUC For population-based validity: IF	Combined categories of SBA as main provider at delivery had a high sensitivity (95.0%) and a low specificity (15.2%), resulting in a low individual-level reporting accuracy (AUC: 0.55, 95% CI: 0.51 to 0.59) and a low population-level bias (IF: 1.02). Three provider categories were used: doctor/medical resident, nurse/midwife and student nurse, of which the first two were considered SBA. Main provider was a doctor/medical resident, had a high individual-level accuracy (AUC: 0.86, 95% CI: 0.83 to 0.89) and a large population-level bias (IF: 1.63). Main provider was a nurse/midwife, had a high individual-level accuracy (AUC: 0.80, 95% CI: 0.76 to 0.83) and a low population-level bias (IF: 0.93). Main provider was a student nurse, had a low individual-level accuracy (AUC: 0.57, 95% CI: 0.53 to 0.61) and a large population-level bias (IF: 0.45). There was a tendency for women’s self-report to misclassify medical residents and nurse/midwives as doctors and to misclassify student nurses as nurse/midwives.
McCarthy et al. 201619	Two public hospitals in Kisumu and Kiambu districts, Kenya. Pregnant women aged 15–44 years admitted to study facilities for labour and delivery and who participated in the baseline interview and were reinterviewed in the community. n=515	Interviews conducted 13–15 months after delivery.	Direct observation by registered nurse/midwives and the woman’s previous exit interview at hospital discharge.18	For individual reporting accuracy: sensitivity, specificity and AUC For population-based validity: IF For individual-level reliability: Agreement between women’s responses at discharge and follow-up using the phi coefficient, which ranges from −1 (perfect disagreement) to 0 (no correlation) to 1 (perfect agreement)	Main provider at delivery was SBA (constructed category of doctor/medical resident or nurse/midwife), had a high sensitivity (91.0%) and a low specificity (18.0%) at 13–15 months follow-up; AUC at follow-up (0.54, 95% CI: 0.50 to 0.59) and IF at follow-up (0.98) were similar to baseline AUC and IF reported by Blanc and colleagues.18 There was some deterioration in individual-level reporting accuracy for main provider at delivery was a doctor/medical resident at 13–15 months follow-up (AUC: 0.77, 95% CI: 0.73 to 0.81) compared with baseline (AUC: 0.86, 95% CI: 0.82 to 0.89) and for a nurse/midwife at follow-up (AUC: 0.70, 95% CI: 0.66 to 0.74) compared with baseline (AUC: 0.80, 95% CI: 0.76 to 0.83). Population-level bias for SBA coverage remained low overall and was very similar between baseline and follow-up (IF: 1.0 vs 0.98). Population-level bias was larger at follow-up compared with baseline for main provider was a doctor/medical resident (IF: 2.44 vs 1.57) and nurse/midwife (IF: 0.76 vs 0.94). Reliability of women’s reports of the main provider during delivery between baseline and 13–15 months follow-up was low (r_phi=0.32) for both doctor/medical resident and nurse/midwife.
Hussein et al. 200420	Two hospitals in the Greater Accra region, Ghana. Women who had delivered in the study facility in the 10 days before interview. n=9	Interviews conducted with women up to 10 days after delivery.	Birth register and clinical notes of the delivery; interviewers also asked health personnel to recollect circumstances of the birth.	Not assessed.	In seven of nine cases, respondents identified their main attendant as was recorded in the birth register. Of the two discordant cases, the respondents reported delivering without an attendant or that the midwife arrived after delivery of the baby’s head. In both cases, the register recorded the birth assisted by a midwife with no mention of partial or non-attendance.

*Plots the indicator’s sensitivity (‘true positive’) against its false positive rate (1-specificity). AUC values range from 0 (zero accuracy) to 1.0 (perfect accuracy) with a value of 0.5 being the equivalent of a random guess.
†Ratio of the prevalence as self-reported by women over the ‘true prevalence’ according to the gold standard comparison. An IF of 1.0 indicates no bias.
AUC, area under the receiver operating characteristic curve; IF, inflationfactor; SBA, skilled birth attendant.