Concordance between GPS-based smartphone app for continuous location tracking and mother's recall of care-seeking for child illness in India

Linking household surveys and facility assessments: a comparison of geospatial methods using nationally representative data from Malawi

BACKGROUND

Linking facility and household surveys through geographic methods is a popular technique to draw conclusions about the relationship between health services and population health outcomes at local levels. These methods are useful tools for measuring effective coverage and tracking progress towards Universal Health Coverage, but are understudied. This paper compares the appropriateness of several geospatial methods used for linking individuals (within displaced survey cluster locations) to their source of family planning (at undisplaced health facilities) at a national level.

METHODS

In Malawi, geographic methods linked a population health survey, rural clusters from the Woman's Questionnaire of the 2015 Malawi Demographic and Health Survey (MDHS 2015), to Malawi's national health facility census to understand the service environment where women receive family planning services. Individuals from MDHS 2015 clusters were linked to health facilities through four geographic methods: (i) closest facility, (ii) buffer (5 km), (iii) administrative boundary, and (iv) a newly described theoretical catchment area method. Results were compared across metrics to assess the number of unlinked clusters (data lost), the number of linkages per cluster (precision of linkage), and the number of women linked to their last source of modern contraceptive (appropriateness of linkage).

RESULTS

The closest facility and administrative boundary methods linked every cluster to at least one facility, while the 5-km buffer method left 288 clusters (35.3%) unlinked. The theoretical catchment area method linked all but one cluster to at least one facility (99.9% linked). Closest facility, 5-km buffer, administrative boundary, and catchment methods linked clusters to 1.0, 1.4, 21.1, and 3.3 facilities on average, respectively. Overall, the closest facility, 5-km buffer, administrative boundary, and catchment methods appropriately linked 64.8%, 51.9%, 97.5%, and 88.9% of women to their last source of modern contraceptive, respectively.

CONCLUSIONS

Of the methods studied, the theoretical catchment area linking method loses a marginal amount of population data, links clusters to a relatively low number of facilities, and maintains a high level of appropriate linkages. This linking method is demonstrated at scale and can be used to link individuals to qualities of their service environments and better understand the pathways through which interventions impact health.

Determinants and patterns of care-seeking for childhood illness in rural Pune District, India

BACKGROUND

An estimated 1.2 million children under five years of age die each year in India, with pneumonia and diarrhea among the leading causes. Increasing care-seeking is important to reduce mortality and morbidity from these causes. This paper explores the determinants and patterns of care-seeking for childhood illness in rural Pune district, India.

METHODS

Mothers having at least one child <5 years from the study area of the Vadu Health and Demographic Surveillance System were enrolled in a prospective cohort study. Household sociodemographic information was collected through a baseline questionnaire administered at enrollment. Participants were visited up to six times between July 2015 and February 2016 to collect information on recent childhood acute illness and associated care-seeking behavior. Multivariate logistic regression explored the associations between care-seeking and child, participant, and household characteristics.

RESULTS

We enrolled 743 mothers with 1066 eligible children, completing 2585 follow-up interviews (90% completion). Overall acute illness prevalence in children was 26% with care sought from a health facility during 71% of episodes. Multivariable logistic regression showed care-seeking was associated with the number of reported symptoms (Odds ratio (OR) = 2.4, 95% confidence interval (CI) = 1.5-3.9) and household insurance coverage (OR = 2.2, 95% CI = 1.1-4.3). We observed an interaction between the associations of illness severity and maternal employment on care-seeking. Somewhat-to-very severe illness was associated with increased care-seeking among both employed (OR = 5.0, 95% CI = 2.2-11.1) and currently unemployed mothers (OR = 7.0, 95% CI = 3.9-12.6). Maternal employment was associated with reduced care-seeking for non-severe illness (OR = 0.3, 95% CI = 0.1-0.7), but not associated with care-seeking for somewhat-to-very severe illness. Child sex was not associated with care-seeking.

CONCLUSIONS

This study demonstrates the importance of illness characteristics in determining facility-based care-seeking while also suggesting that maternal employment resulted in decreased care-seeking among non-severe illness episodes. The nature of the association between maternal employment and care-seeking is unclear and should be explored through additional studies. Similarly, the absence of male bias in care-seeking should be examined to assess for potential bias at other stages in the management of childhood illness.

Validating a GPS-based approach to detect health facility visits against maternal response to prompted recall survey

Introduction

Common approaches to measure health behaviors rely on participant responses and are subject to bias. Technology-based alternatives, particularly using GPS, address these biases while opening new channels for research. This study describes the development and implementation of a GPS-based approach to detect health facility visits in rural Pune district, India.

Methods

Participants were mothers of under-five year old children within the Vadu Demographic Surveillance area. Participants received GPS-enabled smartphones pre-installed with a location-aware application to continuously record and transmit participant location data to a central server. Data were analyzed to identify health facility visits according to a parameter-based approach, optimal thresholds of which were calibrated through a simulation exercise. Lists of GPS-detected health facility visits were generated at each of six follow-up home visits and reviewed with participants through prompted recall survey, confirming visits which were correctly identified. Detected visits were analyzed using logistic regression to explore factors associated with the identification of false positive GPS-detected visits.

Results

We enrolled 200 participants and completed 1098 follow-up visits over the six-month study period. Prompted recall surveys were completed for 694 follow-up visits with one or more GPS-detected health facility visits. While the approach performed well during calibration (positive predictive value (PPV) 78%), performance was poor when applied to participant data. Only 440 of 22 251 detected visits were confirmed (PPV 2%). False positives increased as participants spent more time in areas of high health facility density (odds ratio (OR) = 2.29, 95% confidence interval (CI) = 1.62-3.25). Visits detected at facilities other than hospitals and clinics were also more likely to be false positives (OR = 2.78, 95% CI = 1.65-4.67) as were visits detected to facilities nearby participant homes, with the likelihood decreasing as distance increased (OR = 0.89, 95% CI = 0.82-0.97). Visit duration was not associated with confirmation status.

Conclusions

The optimal parameter combination for health facility visits simulated by field workers substantially overestimated health visits from participant GPS data. This study provides useful insights into the challenges in detecting health facility visits where providers are numerous, highly clustered within urban centers and located near residential areas of the population which they serve.

Assessing the reactivity to mobile phones and repeated surveys on reported care-seeking for common childhood illnesses in rural India

Background

Traditionally, health care-seeking for child illness is assessed through population-based and nationally representative demographic and health surveys (DHS) that are conducted once every five to seven years and are based on maternal recall. These maternal reports are subject to recall bias. Mobile phones (with the use of GPS technology) have the potential to constantly track movements of phone owners and provide high quality and more accurate data at a population level in low and middle income countries (LMICs) to assess the validity of maternal recall. We provided a group of mothers with smartphones installed with a location-aware application and visited them monthly to administer a survey questionnaire on care-seeking for diarrhoea, fever and cough with fever. This paper assesses for any reactivity to smartphones or repeated study contacts for measuring care-seeking and if this resulted in change in health care provider preference.

Methods

We enrolled 749 mothers from rural areas of Pune district in Maharashtra, India and randomly allocated them to one of three groups – a longitudinal phone group, a longitudinal control group and a cross-sectional control group. We collected baseline information from mothers, including individual and household demographic and socio-economic characteristics and care-seeking preferences for child illness. We followed up both longitudinal groups monthly and each cross-sectional sub-group once over a period of 6 months. At each follow up, we administered questions identical to those in the National Family Health Survey (NFHS) questionnaire to determine an episode of diarrhoea, fever or cough within the last 15 days, care seeking for the same, and the type of provider. The data were analysed using the χ² test or Fisher Exact Test for categorical variables, or with the Kruskall-Wallis non-parametric test for continuous variables (due to the non-normal nature of the data). Multivariable joint models of group and visit time were analysed with logistic regression methods.

Results

All three groups were similar in their socio-demographic characteristics at baseline. We did not observe any significant difference in care seeking for diarrhoea, fever or cough with fever between groups. Also, we did not observe any significant difference in proportion of children seeking care from the private sector.

Conclusions

We did not observe any reactivity in this study due to the presence of the phone (Hawthorne effect) or repeated study visits. The study also shows the potential of using GPS enabled smartphones to enrich DHS surveys in LMICs like India. However, further studies need to be conducted in other population groups before the findings can be generalised.