Dupont 2017.

Methods	Cluster randomised trial OPERA trial
Participants	Births at 95 maternity units in France; 50% had 500–1499 deliveries per year, 40% had > 1499 deliveries per year, and 10% had < 500 deliveries per year.
Interventions	Intervention: single limited complex intervention to promote national guidelines on antenatal care and peripartum practices; perinatal MMCs in presence or absence of a CP. The first component was an outreach visit to brief obstetricians, midwives and anaesthetists on the national guidelines on morbidity/mortality case management. Precisely, the guidelines were about monitoring of normal pregnancy, management of intrauterine growth‐restricted fetuses during labour, interpretation of fetal heart rate during labour and safe practice in instrumental vaginal delivery. During the visit, the co‐ordinators discussed the scientific and medical validity of the guidelines with the medical staff to support their implementation. The second component was a series of MMCs dedicated to review a selected sample of perinatal morbidity and mortality cases with all staff members who managed them. The MMCs were held in the maternity units concerned and were led by the investigators, who invited all the staff to attend. MMC + CP sessions included analyses of the staff decision‐making processes and explored the role of psychological factors in these processes. 3–4 MMCs were held in maternity units with > 1500 deliveries, 2–3 in units with 500–1500 deliveries, and 2 in units with < 500 deliveries per year. Control group: continued current care – no intervention.
Outcomes	Primary outcome Rate of suboptimal care among cases of perinatal morbidity or mortality (stillbirths and neonatal). [Optimal care was defined as the management of a morbidity or mortality case in perfect accordance with the guidelines. In each case, the quality of care was jointly examined by two reviewers who had to agree on classifying it according to a 7‐point scale, where 7 = optimal, 6 = nearly optimal, 5 = satisfactory, 4 = nearly satisfactory, 3 = possibly suboptimal, 2 = certainly suboptimal, and 1 = not classifiable. Suboptimal care was defined as score 3 or 2 (i.e. not compliant with the guidelines).] Secondary outcomes Rate of suboptimal care among morbidity cases. Rate of suboptimal care among mortality cases. Rate of avoidable morbidity cases. Rate of avoidable mortality cases. Incidence of morbidity (excluding intrauterine growth restriction). Incidence of mortality.
Notes	Maternal mortality and morbidity were not evaluated. Funded by the French Ministry of Health under its Clinical Research Hospital Program (contract no. 27‐41).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated randomisation was stratified by network, institutional status (public university‐related, public university‐unrelated and private) and annual number of deliveries (< 500, 500–1499 and 1500). A second randomisation within the intervention group generated two subgroups assigned to have MMCs in presence or absence of a CP.
Allocation concealment (selection bias)	Low risk	The health units could not know in advance to which group they would be randomised (Dupont, personal communication).
Blinding of participants and personnel (performance bias) All outcomes	High risk	Impossible to blind participants and personnel. Personnel may behave differently if they are aware that they are receiving an intervention.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	The investigators who evaluated the outcome on suboptimal care were blinded as to whether the case was from an intervention or control site. The same investigators took part in mortality meetings at intervention sites but only a small number of cases were discussed and the outcome evaluation was conducted 1 year later so they could not remember the cases (Dupont, personal communication).
Incomplete outcome data (attrition bias) All outcomes	Low risk	2 private units withdrew (prior to randomisation). All 95 units completed the trial.
Selective reporting (reporting bias)	Low risk	There were no outcomes in the protocol which were not reported in the publication (Dupont, personal communication).
Other bias	High risk	Six units randomised to the intervention group did not implement the intervention and were transferred to the control group and analysed per protocol instead of intention to treat. It is impossible to know whether the transferred units are completely similar to those that retained their original randomisation.
Contamination	Low risk	Although it is possible that an individual health worker could work in both an intervention and a control hospital, it is very unlikely that this would apply to the whole team.

CP: clinical psychologist; MMC: morbidity/mortality conference; OR: odds ratio.