A total of 335 rural household (566 children) and 469 urban households (590 children) were enrolled in the study. At follow-up, 43 households (5.3%) were unavailable to complete the survey because the participating mother(s) had moved outside of the study area or to an unknown residence for the remainder of the study period. At follow-up, 14 households withdrew consent for the follow-up survey. This resulted in a loss-to-follow-up of 7.1% of households.

Characteristics of participating children, mothers, and households are shown in Table 1 . There was an approximately equal distribution of children by age and gender. There were slightly fewer children under one year of age due to the lag period between enrollment and the follow-up survey, which excluded neonates born within the follow-up period. The mean age of mothers was 29.6 years in the rural area and 27.1 years in the urban area. Mothers in the urban area on average had slightly higher education compared to mothers in the rural area.

All government health centers, health posts, and CBAs trained in child curative services within the study area agreed to participate in care-seeking event tracking ( Table 1 ). There were no private facilities or pharmacies within the rural study area. In the urban area, four private clinics and five pharmacies participated in the study while one private clinic and one pharmacy refused to participate. In the rural area, 29 traditional or faith-based practitioners participated in the event tracking. Two churches and one traditional birth attendant in the urban area participated in the event tracking.

No mobile clinics, mission facilities, or private community-based agents were present in the study area. Choma General Hospital was located in the urban study area but excluded from the study due to low anticipated numbers of study participants seeking care at a referral facility and potential disruption caused by event tracking in a high-volume referral facility. A large number of informal shops that stocked unregulated analgesics in addition to grocery items were identified. None were included in the study because they did not receive a significant volume of care-seeking events; seeking care from a shop was reported for 1% of rural children and 5% of urban children with illness (data not shown).

Among the 1083 children included in the care-seeking survey, 34.5% of urban children and 36.4% of rural children experienced at least one illness according to DHS questions (“DHS illness”) in the 2 weeks preceding the survey ( Table 2 ). Fever was the most commonly experienced symptom in both the rural and urban areas. Among those children that experienced a DHS illness, mothers reported care was sought for 78.9% of rural children and 66.5% of urban children. Reported care-seeking from more than one source was uncommon. Among those children taken for care, mothers most often reported their child was taken to a skilled provider (95.8% rural reported care-seeking events, 91.4% urban reported care-seeking events). Overall, 20 care-seeking events (3% of rural and 12% of urban reported care-seeking events) were reported with non-participating providers, including shops (11) and the general hospital (5) (Table S1 in Online Supplementary Document(Online Supplementary Document) ). Some mothers misclassified seeking care from a government hospital or mission clinic when care had actually been sought from a government health center based on name of the facility, including 11% of rural and 1% of urban reported care-seeking events.

Maternal report of care-seeking for child illness was compared against provider-documented care-seeking events. Table 3 presents the distribution of reported and documented (true positive), reported but undocumented (false positive), and unreported but documented (false negative) care-seeking events by provider type and strata among providers participating in event tracking. The majority of care-seeking events occurred at government health centers. CBAs accounted for a significant proportion of rural care-seeking events. A moderate number of mothers reported seeking care from public sector providers with no documentation these care-seeking events occurred. Few mothers reported seeking care in the private sector. A relatively high proportion of documented care-seeking events among traditional practitioners were unreported by mothers.

Maternal report of correct source of care by provider category is presented in Table 4 . Maternal report of correct source of care had a sensitivity, or proportion of true care-seeking events that were correctly reported, of 91.2% (95% CI 83.6-95.5%) in the rural strata and 97.9% (95% CI 92.0-99.5%) in the urban strata. The specificity of maternal report of source of care was lower at 71.4% (95% CI 57.4-82.3%) in the rural and 75.5% (95% CI 62.1-85.3%) in the urban strata. The positive predictive value (PPV), or proportion of reported care-seeking events that truly occurred, was 91.3% (95% CI 85.9-94.8%) and 82.3% (95% CI 74.1-88.3%) in the rural and urban strata respectively. The individual accuracy of maternal report of source of care, calculated as area under the receiver operating characteristic curve (AUC), was 81.1% (95% CI 74.3-87.9%) in the rural strata and 86.6% (95% CI 81.7-91.5%) in the urban strata.

Maternal report of any care-seeking event had a slightly higher sensitivity and specificity than maternal report of source of care by provider category ( Table 4 ). The AUC of maternal report of any care-seeking was also higher at 84.1% (95% CI 77.5-90.6%) in the rural strata and 87.6% (95% CI 82.8-92.4%) in the urban strata.

Maternal report of care-seeking from a skilled provider had the highest sensitivity, specificity, and accuracy of the three report measures ( Table 4 ). The AUC of maternal report of care-seeking from a skilled provider was high at 86.4% (95% CI 80.6-92.3%) and 89.1% (95% CI 84.7-93.5%) in the rural and urban strata, respectively.

The sensitivity, specificity, and accuracy of maternal report of care-seeking by illness type are presented in Table S2 in Online Supplementary Document(Online Supplementary Document) . Estimates of sensitivity and accuracy were similar across illness types, but the specificity of maternal report was lower for mothers of children with fever and/or ARI compared to children with diarrhea. However, the overall precision of illness-specific estimates was limited by low sample size when stratified by illness.

The relationship between accuracy of maternal report of care-seeking by source of care and characteristics of the child, mother, household, illness episode, and true care-seeking behavior / source of care are presented in Table 5 . The association between these characteristics and accuracy of maternal report of any care-seeking and skilled care are presented in Table S3 in Online Supplementary Document(Online Supplementary Document) . There were no significant associations between accuracy of maternal report and characteristics of the child, mother, or household. Adjusting for other characteristics, mothers of children who were not taken for care (AOR = 0.04; 95% CI 0.01-0.11) had significantly lower odds of accurately reporting their care-seeking behavior (no care sought) compared to mothers who sought care from a government health center ( Table 5 ). Mothers of children for whom no care was sought had lower odds of accurately reporting seeking any care and seeking skilled care (Table S3 in Online Supplementary Document(Online Supplementary Document) ). Mothers that sought care from a traditional practitioner had significantly lower odds (AOR = 0.01; 95% CI 0.00-0.05) of accurately reporting the care-seeking event when adjusting for other characteristics ( Table 5 ). Mother that only sought care from traditional practitioners also had significantly lower odds of accurately reporting any care-seeking (Table S3 in Online Supplementary Document(Online Supplementary Document) ). Mothers of children with fever and symptoms of ARI had significantly lower odds (AOR = 0.2; 95% CI 0.05-0.88) of accurately reporting care-seeking events compared to mothers of children with fever alone. However, this association was not significant when assessing characteristics associated with accuracy of reporting any care-seeking or seeking care from a skilled provider.

The modeled coverage of care-seeking by source of care ascertained from maternal report over a range of coverage scenarios demonstrates that at low coverage levels the lower specificity of maternal report would result in substantial overestimation of the proportion of sick children taken for care at the reported source of care ( Figure 1 ). At high coverage levels, the high sensitivity of maternal report would result in only a slight underestimation of proportion of sick children taken for care at the reported source of care. This trend was true for estimation of any care-seeking event and seeking care from a skilled provider. In both the rural and urban strata, all three measures of maternal report had small inflation factors slightly overestimating or producing estimates of care-seeking behavior very close to the true coverage in the study population ( Table 6 ). This was also true of coverage estimated by illness type (Table S4 in Online Supplementary Document(Online Supplementary Document) ).

Figure 1.  Measured vs true care-seeking behavior estimated through maternal report of care-seeking for child illness, by strata.

Reclassification of providers and exclusion of care-seeking events with non-participating providers did not significantly alter estimates of the sensitivity and specificity of maternal report (Tables S5-S6 in Online Supplementary Document(Online Supplementary Document) ). Estimated sensitivity of maternal report of care-seeking by provider category in the rural area was non-significantly lower without reclassification (83%) compared to with reclassification (91%). Estimated specificity of maternal report of care-seeking by provider category in the urban area was non-significantly lower without excluding non-participating providers (64%) compared to with exclusion (76%). There was no evidence that the care-seeking event tracking methods affected maternal report of care-seeking (Table S7 in Online Supplementary Document(Online Supplementary Document) ).

The validity of maternal report of care-seeking for child illness was assessed by comparing maternal report against provider-documented care-seeking events. This study found high sensitivity and reasonable specificity of maternal report of care-seeking for child illness. There have been no other studies of the validity of maternal report of care-seeking for child illness in LMICs. Results from a similar study in Pune, India are forthcoming. A number of studies in high-income countries have assessed accuracy of self-reported adult health service use and found moderate to high agreement with medical records [19-23].

The observed sensitivity of maternal report of source of care was high overall but somewhat lower in rural areas, compared to the urban areas, due in large part to under-reporting of care sought from traditional practitioners. Traditional practitioners were the most common type of health provider in the rural study area, although they saw a relatively low volume of sick children. Within the study area, some local leaders and health workers openly discouraged the use of traditional practitioners, although a handful of traditional practitioners had been incorporated into the public sector as CBAs or members of local health committees or safe motherhood action groups. Concern over negative perceptions of treatment by a traditional practitioner has been cited as a potential cause of under-reporting of the use of traditional practitioners in some settings and may explain the underreporting in this setting [24,25]. Alternatively, mothers may not consider treatment by traditional practitioners to be seeking care, a point that could be clarified in the survey question administration.

The lower specificity of maternal report was driven by over-reporting of seeking care from public sector providers, including government health centers, posts, and CBAs. This over-reporting of care-seeking events in the public sector may be attributable to mothers’ expectation that researchers want to hear that care was sought for a sick child, and greater approval of treatment from a government provider, resulting in a social desirability bias [8].

The study findings were limited by the low diversity in care-seeking practices for child illness and the exclusion of shops. The majority of care-seeking events occurred in the public sector. Despite the availability of a number of private clinics and pharmacies in the urban area, very few care-seeking events were reported or documented in the private sector. Provision of free treatment for children aged under 5 in the public sector may account for low care-seeking in the private sector. The public sector is the primary source of care for child illness in many sub-Saharan African contexts [26]. However, this provider landscape may not be representative in urban areas or in other regions, limiting the generalizability of these findings.

The study cannot draw conclusions about the accuracy of maternal report of care-seeking in the private sector. As pharmacists did not provide consultations and adult formulations of medicines can be used to treat children, it was difficult for pharmacists to know when care was being sought for a child, potentially resulting in under-documentation of care-seeking events. Additionally, care-seeking from shops was not tracked because they did not meet study inclusion criteria. Although reported care-seeking from informal providers was low, absence of tracking data among shops prohibits us from assessing accuracy of maternal report as it relates to these providers.

Provider documentation of care-seeking events through barcode scans and distribution of tokens was imperfect due to issues with keeping the phone charged and accessible, caregivers failing to present the barcode card to providers, providers forgetting to distribute ribbons, and caregiver refusal or loss of ribbons. To account for these limitations, we reviewed health provider records. In the public sector and private clinics, these records were already being kept as part of routine health service tracking and were maintained separately from the study event tracking methods. This independent source of data on treatment of children strengthened the completeness of event tracking data.

While the overall accuracy of maternal report of source of care was high, we found some misclassification of source of care. Based on the originally reported source of care using the ZDHS provider categories, many mothers reported seeking care at a hospital or mission facility when the true source of care was a government health center. Misclassification was most likely due to a number of factors, including use of “hospital” as a colloquial term for health facility and changing authority from mission to government management at one health center in the study area. Misclassification was simple to identify when mothers were asked to report the name of the provider or facility from which care was sought. Inclusion of an additional question on the name of the provider, or additional unrecorded prompting to verify the category of health provider, could reduce misclassification error in household surveys but may not be feasible at scale. Additionally, some provider categories currently used in the ZDHS are non-exclusive and uninformative. CBAs in the study area ran or staffed some government health posts. However, government health posts were categorized with health centers, although the range of service and quality care offered by a CBA compared to a nurse or clinical officer could vary greatly.

This study suggests that maternal report as captured through household surveys is a valid measure of source of care for child illness in settings where utilization of public sector providers is high and diversity of care-seeking is relatively low. This finding is broadly applicable to other setting where the public sector is the primary source of care. However, there is need for additional research to assess the accuracy of maternal report of care-seeking for childhood illness in other contexts, particularly to understand report related to care-seeking in the private formal and informal sectors.