Study Setting.

The study was carried out in low-income households located in Trivandrum, Kerala and a nearby fishing community. Houses were all permanent structures with enclosed indoor bathrooms and toilets. All had piped water, either to the house or a nearby public standpipe. However, water was only supplied for a limited period each day and households stored water in the house for domestic use. Toilets were predominantly pour-flush, connected to sewer systems or septic tanks. All households owned at least one soap bar that was used for personal washing. Soap was rarely stored in toilets when separated from bathrooms, suggesting that HWWS and body washing were not usually practiced in this area.

Sampling.

Household visits by fieldworkers were used to recruit a convenience sample of 30 households in each of the two communities. All households included at least one child, aged between 6 and 11 years, who was attending a Government-aided or Government school. This criterion was included as these households were of broader interest to Unilever’s Lifebuoy brand, which runs hand washing promotion activities in schools.

Loggers.

The loggers used in this study have been described in detail elsewhere [9, 10]. They measure 47mm x 17mm x 14mm and are small enough to be embedded within soap bars. Each logger contains a set of sensors including a 3-dimensional accelerometer, acoustic sensor, tilt sensor, real-time clock and a memory chip. These sensors were electronically calibrated during manufacturing and tested in the lab before use.

When used in trials loggers are inductively charged and programmed using a bespoke hub. During the trial loggers remain in low power mode until activated by movement. Once activated they record the date and time, acceleration, and an acoustic signal within a specified frequency band.

Logged soaps were made by warming a 150g soap bar in a microwave oven. The warmed soap was cut in half and recesses the size of the logger cut in each of the two halves. The logger was then placed inside the recesses and the two halves pressed together using a manual soap press.

Logged water vessels were made by attaching a logger to the inside base of a 1 litre white opaque plastic bottle using waterproof glue.

Outcome measurement.

Data from all households were collected for the purposes of the current study and for validating the loggers’ accuracy at detecting genuine soap use and discriminating HWWS from bathing. Outcomes related to this validation were reported in Zillmer et al. [10]).

Here we estimated the number of ‘HWWS events’ using the raw logger data from the red soaps. First, bespoke algorithms used statistical features based on event duration, sound-level quantile, and acceleration to discriminate between data generated by HWWS and data generated by general movements, unrelated to HWWS. These statistical features were then passed on to a cluster analysis and reference data from laboratory experiments used to identify the cluster that contained HWWS events (see [10] for more details). The number of ‘HWWS events’ was then calculated by combining HWWS occasions separated by less than 60 seconds. This avoids over-counting events due to ‘quiet’ periods in washing routines when soap is not used.

The logged water vessels were used as the indicator of toilet use. The algorithm applied to the raw data identified movement sequences that incorporated inversion of the vessels. This was assumed to indicate the use of the vessel for anal cleansing or flushing of the toilet.

Two main outcome measures were calculated in this study, ‘total HWWS events per household per study day’ (from 1 red bar in the soap in bathroom arm, and 2 red bars in the soap by toilet arm) and the ‘probability of HWWS events post-toilet’. The latter was defined as the per-household proportion of toilet-uses that were associated with a HWWS event. We considered toilet uses and HWWS events to be associated when they were within 2 minutes of each other.

In addition, we calculated the number of ‘white soap events per household per day’ and ‘probability of white soap use post-toilet’ to establish whether the presence of soap by the toilet affected the use of this soap. The use of the white soaps were most likely to reflect body washing but could include hand washing where instructions were not followed. As with the red soap loggers, data from the white soap loggers were analysed using algorithms and events within 60 seconds of each other combined into single events.

Ethics and consenting.

All householders were shown a logger. It was explained to householders that these had been inserted into the study soaps and water vessels to record the times when these items were moved. Informed consent forms were written in malayalam and were signed by the head of the household. Consent forms were read to signees when they could not read.

A payment of 150 INR was made to heads of households on completion of the study. The study protocol was approved by the ethics committee of The London School of Hygiene & Tropical Medicine. Local clearances were obtained by the Socio Economic Unit Foundation, Trivandrum, who also assisted with the field work.

Data management and analysis.

After signal analyses, event data were transferred to JMP for analysis where T-tests were used to compare the two study arms. In order to perform t-tests on the ‘probability of HWWS using after toilet use’ and ‘probability of using white soap after toilet use’, we transformed the raw probabilities using an ‘Empirical Logit’ transform, where [11]. The ‘Empirical logit’ is similar to a Logit transform but uses modified probabilities, pe to remove zero values. pe = (n+0.5)/(N+1).