In a cRCT across 48 urban slum communities of Mumbai, it was possible to implement women's groups in challenging conditions. We did not, however, see substantial effects on health care. Women in intervention clusters reported fewer sentinel antenatal morbidities, but we lack a conceptual basis to explain this. The lower stillbirth rate and higher neonatal mortality rate in the intervention arm have, we think, three possible explanations. The first is chance, or residual confounding: it is possible that intervention and control groups differed systematically in ways unaccounted for by our analysis. Second, the intervention may have reduced stillbirths by encouraging timely care-seeking. It is conceivable that some of these infants did not survive subsequently, shifting the balance of mortality from stillbirth to neonatal death. If this was the case, we might have expected to see fewer fresh stillbirths and more early neonatal asphyxial deaths in the intervention arm, a finding not supported by the verbal autopsy data (unpublished data). Third, the data collection may have itself led to improvements in both trial arms. Given that it consisted of birth identification and an interview at 6 wk postpartum, we think this unlikely.
Municipal figures for the same period confirm a reduction in mortality across the city. The positive changes seen in both intervention and control groups are intriguing. We considered the possibility that external initiatives might have affected intervention and control groups, differentially or equally. Significant municipal initiatives in the trial period included some improvement in outreach services by community health volunteers, birth registration and pulse polio campaigns, and infectious disease surveillance. We have cluster-specific information on concurrent non-governmental initiatives. Two NGOs were working on general health in some clusters. We do not think that specific initiatives explain our findings. More likely, in our opinion, was a general improvement in environmental conditions accompanied by behaviour change. Conditions in slum areas improved manifestly over the trial period. Gutters were covered, sanitation block coverage increased, housing fabric became more durable, and there was widespread electricity supply. We think that health-related culture change is a natural accompaniment, all the more because of aspiration and the notions of modernity of Mumbai's residents.
The surveillance and intervention teams were separate. Each of the six municipal wards had a surveillance supervisor and two investigators, responsible for data collection in all eight clusters: a mix of intervention and control. The procedures were identical in intervention and control clusters, and the supervisors and investigators saw their work as unconcerned with the intervention. It is conceivable that in intervention areas sakhis could have told identifiers about births and deaths. We have discussed this with the field and data management teams, and we do not think that this happened. As local residents, the birth and death identifiers were aware that there was an intervention in their community, but were focused on their task and did not dwell on the comparative nature of the trial.
The trial demonstrated the value of a counterfactual control group and the potential weakness of ecological evaluation, the commonest example being a before-after comparison. If we had based our assessment on the trends in the intervention arm seen in and (a design used by many programs), it would have appeared an unqualified success 
. A fall in documented births in the third year of the trial was partly explained by demolition of some settlements, and by difficulties in follow-up. This illustrates a key limitation of urban initiatives: population mobility and the fact that, the poorer the target group, the more transient are their homes. Public health trials would benefit from census data and better registries. Good registration would certainly help when outcome numbers are small and a single missed stillbirth has a substantial effect on a rate per thousand. Our subsequent trial will use censuses rather than prospective ascertainment.
We think that the trial raises three general issues: coverage, target group, and the emergent pattern of health care in cities in the South. Achieving sufficient intervention coverage has been a challenge in other settings 
. In a situation of space and time constraint, with a lack of social cohesion despite high population density, we did not manage to trigger diffusion of innovation. Using population estimates for Mumbai slum areas from the most recent National Family Health Survey (4.7 members per household; women aged 15–49 constituting 26.7% of the population) 
, our women's groups at their peak involved 8%, and at their nadir 2%, of women of reproductive age (although, as mentioned earlier, outreach communication may have multiplied these figures by up to six times).
Convening community groups was feasible and learning and behaviour change possible, but achieving the impetus necessary for wider change was challenging: group members helped others individually but balked at collective strategizing. There was attrition in group numbers over the course of the intervention, suggesting that women stopped attending when they felt that they had either learned enough or were required to invest more time and energy. Collective action was clearly a big step for groups to take, possible challenges being time-poverty and restrictions on movement, concerns about tenure, lack of confidence, and perhaps a lack of conviction in perinatal health as a major issue. Our target group were slum dwellers, but not exclusively the poorest among them. In a city in which more than half of the population live in slums, slum households themselves encompass a spectrum of socioeconomic realities 
. For potential replicability, the model included municipal wards with a range of infant mortality rates and, although restricted to slum populations, the intervention may not have succeeded in mobilizing the poorest and most at risk, who tend to be hardest to reach. Although socioeconomic status was not associated with differences in trial outcomes between intervention and control arms, our other research in the same population has demonstrated associations between dimensions of vulnerability and maternal and newborn health risks 
and we have described inequities in access to routine maternal health services 
and morbidity care 
. On the basis of these findings, our strategy has changed: subsequent interventions will target the most vulnerable families and we will intensify our efforts to improve quality of care in private and public facilities.
The third issue was the complexity of urban health care. Antenatal care was the norm and the nadir for institutional delivery in trial clusters was 75%. Around 57% of antenatal care and 30% of deliveries were in the private sector (this in a slum-dwelling population). Open access to private providers, and to institutions at all levels of the public sector hierarchy, is a challenge to systematic health care delivery. Our findings confirmed the tendency to bypass public maternity homes, which should handle uncomplicated deliveries, in favour of tertiary institutions. Women's group discussions included clarification of appropriate sites of consultation and considerations of price and quality, reflected in higher use of public sector services by group members. The study underlined the need to work on quality of care in both public and private sectors. Although regulatory insufficiencies make intervention difficult, quality control in the private sector needs to feature more in debates about health care in low-income countries.
The wider implications of our findings include a tipping of the balance of perinatal intervention in cities toward improvement in service quality, with an emphasis on intrapartum vigilance and resuscitation. Indeed, there have been concerns about the utility of cash incentives for institutional delivery in the urban context, the argument being that access is not the primary issue. Our question was not whether women's groups were beneficial to their members. Members valued the groups and their opportunities for peer learning, showed behaviour change, and helped other women in their communities. Exchange of knowledge about health and health services, rights, social networks, and increased confidence are public goods, although there are challenges in quantifying such outcomes in public health terms. Rather, the question was about the added value of women's groups—over and above activities to improve health care quality—in terms of measurable changes in perinatal health at population level. While acknowledging the possibility that others might be able to achieve this through more intensive community activities in higher mortality settings, our own programme did not show effect. Community groups will feature in our subsequent interventions, as they must in any participatory initiative. We will, however, attempt to integrate them more strongly with pro-poorest targeting, service provision at household level, strengthening of links between communities and service providers, and partnerships with public and private sector providers to improve quality of care.