Menu
Menu
Menu
Menu
Background Proactive community case management (ProCCM) has shown promise to advance goals of universa health coverage (UHC). ProCCM community health workers (CHWs) face operational challenges when pursuing their goal of visiting every household in their service area at least twice monthly to proactively find sick patients. We developed a software extension (UHC Mode) to an existing CHW mobile application featuring user interface design improvements to support CHWs in planning daily home visits. We evaluated the effect of UHC Mode on minimum expected home visit coverage. Methods We conducted a parallel-group, two-arm randomised controlled trial of ProCCM CHWs in two separate regions in Mali. CHWs were randomly assigned to UHC Mode or the standard mobile application (control) with a 1:1 allocation. Randomisation was stratified by health catchment area. CHWs and other programme personnel were not masked to arm allocation. CHWs used their assigned intervention for 4 months. Using a difference-in-differences analysis, we estimated the mean change in minimum expected home visit coverage from preintervention to postintervention between arms. Results Enrolment occurred in January 2019. Of 199 eligible CHWs randomised to the intervention or control arm, 196 were enrolled and 195 were included in the analysis. Households whose CHW used UHC Mode had 2.41 times higher odds of minimum expected home visit coverage compared with households whose CHW used the control (95% CI 1.68 to 3.47; p<0.0005). Minimum expected home visit coverage in the UHC Mode arm increased 13.6 percentage points (95% CI 8.1 to 19.0) compared with the control arm. Conclusion Our findings suggest UHC Mode is an effective tool that can improve home visit coverage and promote progress towards UHC when implemented in the ProCCM context. User interface design of health information systems that supports health workers’ daily practices and meets their requirements can have a positive impact on health worker performance and home visit coverage.
We conducted a parallel-group, two-arm randomised controlled trial in the community setting in two health catchment areas in separate regions in Mali, from August 2018 to July 2019 (see figure 1 for Consolidated Standards of Reporting Trials diagram). Trial profile. CHWs, community health workers; UHC, universal health coverage. CHWs in Tori and Yirimadio providing health services based on the ProCCM model were eligible. Although this study took place in a larger context of ProCCM care delivery, a full description of which is published elsewhere,11 12 it is unrelated to the separate ProCCM trial.12 Tori, a rural area in the Mopti region in central Mali, had a population of 29 029 in 2019, while Yirimadio, a periurban area in Bamako, the nation’s capital, had an estimated population of 176 089. Both areas were each served by one public sector primary health centre at the time of the trial. Three CHWs delivering ProCCM in Yirimadio were randomly selected to pilot test UHC Mode before its launch. CHWs who did not provide written informed consent, or had to drop out of the study before intervention launch, or were involved in pilot testing, were excluded from the trial. Although additional secondary outcomes were originally planned, these were dropped due to operational constraints and unforeseen challenges in study implementation (described in the Data preparation section of online supplemental materials). bmjgh-2021-007205supp001.pdf Eligible CHWs not involved in pilot testing were randomly assigned to the control or UHC Mode arm with a 1:1 allocation using a computer-generated random number. Randomisation was stratified by health catchment area, and performed by a research team member who was not involved in study recruitment. Due to operational constraints, randomisation was done before study recruitment. The nature of UHC Mode precluded CHW participants, CHW supervisors and other study personnel from being masked to arm allocation. To avoid contamination, two training sessions were held before intervention launch: one for all eligible CHWs to discuss standard definitions of households and home visit protocols, and a second for CHWs randomised to UHC Mode, to provide an overview of its features, and discourage discussions about UHC Mode with CHWs in the control arm. During the study period, eligible CHWs followed the standard ProCCM protocol: conduct home visits for a minimum of 2 hours per day, 6 days per week, in order to visit each household in their service area with the minimum expected frequency (at least twice monthly).11 12 During their home visits, CHWs reinforced the importance of rapid care-seeking and encouraged community members to call their mobile telephone number or visit them right away if their child became symptomatic. CHWs were expected to be on call to provide care outside of their regular home visiting hours.12 CHWs were also expected to register and maintain information on all households and patients in their service area and to document each home visit and sick patient evaluation using a CHW application preinstalled on Android phones. CHWs maintained autonomy and flexibility in planning their daily home visit workflows, including what order they visited households. Preintervention (baseline) was defined as August 2018 to March 2019. In the latter half of preintervention (December 2018–March 2019), a CHW Supervision Dashboard was implemented in both arms to enable dedicated CHW supervisors to monitor CHW performance indicators related to timeliness and quantity of monthly home visits and quality of care.15 For 4 months postintervention (April–July 2019), eligible CHWs used their randomised intervention: the standard CHW application (control) or the CHW application with UHC Mode (intervention). UHC Mode was deployed in Yirimadio on 8 March 2019 and in Tori on 13 March 2019. UHC Mode is an add-on (software extension) to the CHW application consisting of user interface design improvements to support CHWs to achieve their minimum expected home visit coverage of each household in their service area each month. It features four main design elements: First, each household’s date of last visit is displayed on individual household profiles, which show households’ primary contact information and registered members. This information is displayed on the household list as the time elapsed (in days or months) since the last home visit and colour coded red (if the date of last visit was 30 or more days ago) or black (if the date of last visit was fewer than 30 days ago). Second, red exclamation point icons appear on the household list to emphasise households with fewer than two visits in the month. Third, a modified traffic light colour scheme (red, orange, light blue) shows households receiving zero, one or two or more visits during the month. Finally, the default ordering of the household list was changed from alphabetical by name to chronological by least recent visit date (figure 2). Display of information on the household list in the standard CHW application (A1–A2) versus UHC Mode (B1–B2) and on individual household profiles in the standard application (A3) versus UHC Mode (B3). A1 shows the household list in the standard CHW application. A2 shows the household list in the standard CHW application, with default alphabetical ordering. B1 shows the household list in UHC Mode, which displays the time elapsed in days since the last home visit (colour coded red if the date of last visit was 30 or more days ago, and black if the date of last visit was fewer than 30 days ago). Red exclamation point icons emphasise households with fewer than two visits in the month, and a modified traffic light colour scheme (red, orange, light blue) show households receiving zero, one, or two or more visits in the month. B2 shows the household list in UHC Mode, with default chronological ordering by date of last home visit. A3 shows an individual household profile in the standard CHW application. B3 shows an individual household profile in UHC Mode, with information about the date of last visit and the monthly home visit count. CHW, community health worker; UHC, universal health coverage. UHC Mode was the result of an iterative, human-centred design process,20 which involved direct observations of CHWs planning their daily home visits, focus groups with CHWs and CHW supervisors to elicit feedback on design features to promote higher minimum expected home visit coverage, and months of iterative pilot testing to incorporate CHW feedback into prototype development. The initial prototype focused on redesigning the patient lists feature, which CHWs had identified in focus groups as a bottleneck to managing their daily home visit workflows. Prototype development was informed by key user experience principles discussed in design literature25: visibility, feedback and positive reinforcement. The aim was to increase the visibility of minimum expected home visit coverage information, to provide users with clear feedback whenever their actions had contributed to the goal of minimum expected home visit coverage, and to provide feedback in a way that positively reinforced practices conducive to high coverage rates. The control was the standard CHW application used by all eligible CHWs before the study, which was built with the open-source Community Health Toolkit.26 Key features include job aids for each care protocol, a task list, a report tab summarising CHW performance related to predefined targets, and a patient list organised by household, with access to patients’ longitudinal records. CHWs can organise their daily workflow using the task management, targets and patient list features, and administer clinical care protocols with job aids at the point of care. The application is designed to work while offline and to sync data to a central server when connectivity is available. Important preconditions to UHC Mode’s successful implementation identified during pilot testing were incorporated into the CHW application by intervention launch and available to CHWs in both arms. These included faster application performance, and new functionality (household ‘muting’ forms) to enable CHWs to deactivate (and reactivate) households that moved from the CHW service area (temporarily or permanently), declined CHW services, or no longer required CHW services for other reasons. Data on home visits were collected using the CHW application preintervention and via the assigned intervention postintervention, and extracted more than 3 months after the study period. We counted a maximum of one visit per unique household per day, to reduce the potential for report duplication. If a CHW recorded more than one visit for the same household on the same day, then only one of those visits was counted. Observations were included in the analysis if households were registered and visits were conducted within the study period, and households were considered active (household still present based on household muting form information) for at least 2 days in the month. Data were prepared and analysed using Stata V.15 (Stata). Additional information on data sources and preparation are given in online supplemental appendix 1. We used a binary primary outcome variable, whether an individual household in the CHW service area had minimum expected home visit coverage (at least two home visits by the CHW) in the month. The primary outcome measure was the mean change in minimum expected home visit coverage (defined as an individual household’s receipt of at least two visits in the month) from preintervention to postintervention between arms. We also calculated the service area coverage of minimum expected home visits (the percentage of households visited at least twice in a month) by arm and the unadjusted difference from preintervention to postintervention between arms. Study sample size was primarily determined by operational constraints; all available ProCCM CHWs in Tori and Yirimadio were eligible for the study. We first performed a descriptive analysis to assess comparability between arms in terms of CHWs’ sociodemographic characteristics, baseline service area coverage, actual household load and the wealth quintile distribution among households in the service area. Calculation of actual household load (the number of active households in the CHW service area) was enabled by the household muting functionality incorporated into the CHW application in both arms by intervention launch. To allow sufficient time for CHWs to implement the functionality, we used the number of active households in the final month of the postintervention period and applied it to all earlier months, assuming this was constant during the study period. We next calculated the unadjusted difference in CHW service area coverage from preintervention to postintervention between arms. Service area coverage was graphed to compare trends by arm and month, and preintervention versus postintervention. Comparisons were done with health catchment areas combined and stratified given that baseline service area coverage varied by health catchment area. To analyse the primary outcome measure and estimate UHC Mode’s effect on CHW coverage of a given household in the service area with minimum expected home visits, we conducted a difference-in-differences analysis using a logit model for panel data with CHW-level random effects: where Y is the log odds of coverage of minimum expected home visits observed for household h by CHW i in health catchment area a in month m; Ti is an indicator variable for a CHW’s arm allocation; Pm is an indicator variable for the postintervention period; Za is an indicator variable for health catchment area; and ehim is the idiosyncratic error term. The β3 coefficient on the interaction term, Ti*Pm, estimates the difference-in-differences in the log odds of minimum expected home visit coverage, comparing the mean change (from preintervention to postintervention) in the log odds of minimum expected home visit coverage in households whose CHW used UHC Mode versus the mean change in households whose CHW used the control. We used the margins command in Stata to estimate the absolute change in the predicted probabilities of minimum expected home visit coverage in the UHC Mode versus the control arm. SEs were clustered by CHW, the unit of randomisation. Sensitivity analyses were performed to check the robustness of our regression model to controlling for: month fixed effects (to account for any systematic variations over time); alternative definitions of the preintervention period (ie, 4 months with the CHW Supervision Dashboard and dropping the transition month when UHC Mode launched); and including zero home visit counts for households indicated by CHWs to be temporarily or permanently inactive (to account for bias from potential under-reporting in the event that CHWs who used UHC Mode were more likely than CHWs who used the control to report inactive households). In additional exploratory analyses, we assessed differences in intervention effects by health catchment area (Tori vs Yirimadio), levels of baseline service area coverage, and levels of actual household load. We also performed a subgroup analysis to evaluate whether UHC Mode improved minimum expected home visit coverage of households in the poorest wealth quintile compared with richer households and potentially improved equity in service delivery. This subgroup analysis was limited to Tori because household wealth information was unavailable for Yirimadio, and programme records showed that households in the poorest wealth quintile in Tori had significantly lower odds of minimum expected home visit coverage compared with households in higher wealth quintiles. For exploratory analyses, we repeated the overall main effects analysis, adding a three-way interaction term including one of the following: a dichotomous variable for health catchment area; a categorical indicator variable for baseline service area coverage quartiles; a categorical indicator variable for actual household load quintiles; or a categorical indicator variable for household wealth quintiles (Tori only). The percentile variables for baseline service area coverage and actual household load were calculated separately by health catchment area. We obtained household wealth quintile information for Tori by conducting a principal component analysis of household survey data on household assets,12 adapted from the Demographic and Health Survey. Design of the intervention was informed by engaging with, and soliciting input from, CHWs and CHW supervisors. Patients and the public were not directly involved in the study design, conduct, reporting or dissemination plans of this research. The two study sites are longstanding operational research sites, where Muso and its partners have worked in close partnership with local communities for years. Muso began operational research to improve access to care, maternal health, and child survival in Yirimadio in 2005, and in Tori in 2016. In each area, Muso works closely with local Ministry of Health officials and community members via the Associations de Santé Communitaire, Community Health Associations. These community partners have authorised Muso to conduct operational research related to ProCCM, and actively engage as partners in this research.
N/A
