Data collection

We used cross-sectional data collected by non-participatory observations performed from May 2016 to August 2016. We collected information on number of admissions for the last three months from the date of our visit.

We used three observation checklist tools adopted from internationally recommended checklists [12–15] and utilizing the tool from the World Health Organization concept of five moments of hand hygiene to observe hand-hygiene in newborn care units and labour rooms [12]. The five moment for hand hygiene as mentioned in WHO guidelines are before touching a patient, before clean or aseptic procedures, after risk of exposure to body fluids, after touching a patient and after touching patient surroundings. Separate tools for labour room and newborn care units were developed on an Android based application, linked to the backend server. We used Android based Lenovo tablets for data collection and upload.

The tools were pilot tested in three facilities: two medical colleges and one area hospital, at three different times: twice when we were yet to transfer the tools in mobile application and once after development of the application. The tools and functionality in the application were modified as per findings from these pilots—we fixed the number of hours for observation, simplified data recording, and incorporated drop down options and quality checks. We appointed nursing graduates as observers and trained them extensively for observations and data extraction from registers. Six teams of 4 members each- one supervisor, one field lead and two observers, one for labour room and one for newborn units, were placed in the allocated facilities for six days. In newborn care units, observers spent a minimum of 4 h and in labour rooms a minimum of 6 h every day in direct observation, as per schedule, in either of the two shifts—morning or evening. We kept a longer period of observation in labour rooms so as to be able to observe at least two to three deliveries.

In newborn care units, we observed every contact of a healthcare provider with the admitted newborn for hand hygiene practice during the observation period. The observation unit was thus a contact with the newborn or their environment. However, the number of contacts depended on the severity of sickness as the very sick may get frequent contacts with the healthcare provider. This may also be linked with level of facility where the tertiary hospitals are likely to receive very sick babies. Thus, we adjusted for potential clustering at the facility level due to this sampling method.

In the labour room, we observed 2–3 mothers for hand hygiene compliance during per-vaginal examinations and before conducting delivery, during the 6 h observation period each day. The observers were instructed to observe only one woman at a time even if there was more than one delivery being conducted simultaneously.

Collected data were saved daily and uploaded on a safe server weekly. The data was extracted in MS Excel and checked on a weekly basis. We maximised data quality using several ways: i) inbuilt skips, ranges and checks in the application ii) supervisory visits by lead researchers iii) daily reporting on the number of observations and iv) three levels of data checking exercise- by field supervisors, research assistants and senior investigator.