Completeness of patient-held records: Observations of the Road-to-Health Booklet from two national facility-based surveys at 6 weeks postpartum, South Africa

Background Continuity of care is important for child well-being in all settings where postnatal retention of mother-infant pairs in care remains a challenge. This analysis reports on completeness of patient-held infant Road to Health Booklets (RtHBs), amongst HIV exposed and unexposed infants during the first two years after the RtHB was launched country-wide in South Africa. Methods Secondary data were analysed from two nationally representative, cross-sectional surveys, conducted in 2011-12 and 2012-13. These surveys aimed to measure early effectiveness of the national programme for preventing vertical HIV transmission. Participants were eligible for this analysis if they were 4-8 weeks old, receiving their six-week immunisation, not needing emergency care and had their RtHBs reviewed. Caregivers were interviewed and data abstracted from RtHBs. RtHB completeness across both surveys was defined as the proportion of RtHBs with any of the following indicators recorded: infant birth weight, BCG immunisation, maternal syphilis results and maternal HIV status. A partial proportional odds logistic regression model was used to identify factors associated with completeness. Survey sampling weights were included in all analyses. Results Data from 10 415 (99.6%) participants in 2011-12 and 9529 (99.2%) in 2012-13 were analysed. Overall, recording of all four indicators increased from 23.1% (95% confidence interval (CI) = 22.2-24.0) in 2011-12 to 43.3% (95% CI = 42.3-44.4) in 2012-13. In multivariable models, expected RtHB completeness (ie, recording all four indicators vs recording of < 4 indicators), was significantly (P < 0.05) associated with survey year, marital status, socio-economic status, maternal antenatal TB screening, antenatal infant feeding counselling, delivery at a clinic or hospital and type of birth attendant. Conclusions Routine patient-held infant health RtHB, a critical tool for continuity of care in high HIV/TB prevalence settings, was poorly completed, with less than 50% of the RtHB showing expected completeness. However, government efforts for improved usage of the booklet were evidenced by the near doubling of completeness from 2011 to 2013. Education about its importance and interventions aiming at optimising its use without violating user privacy should be continued. Data from two nationally-representative cross-sectional South African PMTCT surveys conducted in 2011-12 (August 2011 – March 2012) and 2012-13 (October 2012 – May 2013) were used. These surveys were designed primarily to measure MTCT by 6 weeks postpartum. The surveys enrolled mother/caregiver-infant pairs at public primary health care facilities (PHCs) and community health centres (CHCs) offering immunization services. A multi-stage (at provincial, health facility and individual levels), stratified cluster sampling approach was used. Facility size was measured using expected number of DTP (diphtheria, tetanus, and pertussis) immunisations at 6 weeks. The strata were then created by combining the determined size with known maternal HIV prevalence (categorised into either high or low relative to the national average). Sample size at provincial and facility levels were then determined using probability proportionate to size with a target of providing reasonable provincial and national level estimates. Details of the primary studies have been published elsewhere [11,12]. Ethical approval for the study was granted by the South African Medical Research Council’s ethics committee and provincial research ethics committees. The protocol was also approved by the United States Centers for Disease Control and Prevention (Atlanta Georgia, USA) Center for Global Health Associate Director for Science. In each of the primary studies, trained data collectors recruited eligible mother/caregiver infant pairs consecutively or systematically, depending on facility size, until the targeted sample size was achieved. Infants aged 4-8 weeks who were receiving their six-week immunisation on the day of visit and who were not in need of emergency care were eligible for inclusion in the study following signed informed consent (administered in the participant’s preferred language). Face-to-face interviews were conducted after the routine visit activity was completed and data were captured electronically using mobile phones and then stored in an access-restricted database. Data collected through interviews included maternal socio-demographic backgrounds, antenatal care histories and early postnatal health care uptake. All enrolled participants were asked to present the RtHB during the study interview. During the survey interviews, data were extracted from the RtHB, including infant birth weight, BCG immunization, infant HIV exposure status and maternal syphilis testing results. These data are from the earliest two years after the full country-wide implementation of the RtHB (and phasing out of the RtHC) to which PMTCT indicators had been added. The primary outcome used is “completeness of the RtHB”. Therefore, the data for this study were restricted to the sample of enrolled participants who brought the RtHB during the study interview and these were N = 10 415 (99.6%) in 2011-12 and N = 9529 (99.2%) in 2012-13. We created a composite outcome variable using extracted data for four variables that should have been completed at birth, namely; infant birth weight, BCG immunization, maternal HIV status and indication of whether maternal syphilis testing was done. Therefore, the outcome variable is ordinal with counts from 0 through to 4. These indicators were simply chosen because they were of common interest to both survey aims and are also important to assess PMTCT and maternal and child health service uptake. Maternal characteristics potentially associated with uptake of health care services were assessed for association with completeness of the RtHB. These were age, educational qualification, marital status, parity, knowledge of MTCT modes and relative socioeconomic status (SES). Participants were defined as having good knowledge of MTCT if they understood the definition of MTCT and could correctly identify all three modes of MTCT viz. transmission during pregnancy, childbirth and breastfeeding, whilst those who did not understand the definition of MTCT or were not able to identify all modes were defined as having poor knowledge of MTCT. SES, grouped into quartiles, was calculated for each year using principal component analyses from household characteristics (which included type of housing, sanitation, water and fuel), household possessions (such as TV, stove, radio), any food shortage and source of income [13]. Differences between provinces and survey years were also evaluated in relation to the primary outcome. Variables which indirectly reflect competence of health service provision before 6 weeks postpartum were also included to give an indication of whether incompleteness of the booklet is related to the service provider. These variables were receipt of TB screening during pregnancy, receipt of infant feeding counselling during pregnancy, place of delivery (hospital, clinic or home) and type of birth attendant (doctor, nurse/midwife/community health worker or traditional birth attendants). We hypothesize that performance of health care providers in providing basic services recommended during pregnancy, would reflect their diligence is recording patient-held records, including the RtHB when it is issued. Data analysis was done in STATA version SE 13 (Stata Corp, College Station, TX, USA). An ordered logistic regression analyses was used for the outcome variables. Survey sampling weights were used to account for the sample design (ie, multi-stage & strata size proportionality) and realisation (ie, adjustment for attained vs target sample size within each stratum). An additional subcategory was created for independent variables which had more than 5% of data entries in each survey year missing or with ‘don’t know’ responses to the questionnaire. This was done to minimise deviation from the correct estimates when applying the survey sampling weights which were calculated based on the actual attained sample size. The sub-category for ‘unknown’ responses was created for TB screening during delivery and for knowledge of MTCT modes. Factors associated with completeness of the RtHB were identified using a partial proportional odds (PPO) logistic regression analyses in three steps [14,15]. The assumption of proportionality of odds across sub-group pairs of the outcome variable, made by the proportional odds logistic regression model, was first tested for all predictor variables using the Brant test. The Brant test output for proportionality of odds across the categories of the outcome variable for each predictor variable are presented as χ2 statistics and probability values for the null hypothesis that the odds of a predictor variable are proportional across the different binary groupings of the outcome variable (see Table S1 in Online Supplementary Document(Online Supplementary Document)). The possible ordered binary groupings were; completeness ≥1 vs <1, completeness ≥2 vs <2, completeness ≥3 vs <3 and completeness = 4 vs <4 recorded variables. The proportionality of odds assumption was supported in four predictor variables (mother’s education, marital status, parity and birth attendant) and hence these were constrained accordingly in the regression analyses. The rest of the predictor variables violated the proportional odds model assumption and thus were not constrained under this assumption. The PPO model was then used with a function to only constrain proportionality to those predictor variables which were consistent with the parallel regression assumptions (Brant test p-values ≥0.05) and leaving the rest unconstrained. In step 1, separate bivariate PPO logistic regression analyses taking into account the proportionality test results and constraining/not constraining accordingly, were done between the outcome and each predictor variable (estimates ‘unadjusted’ for possible confounding by other variables). In step 2, those with Wald’s test P-value <0.25 in step 1, indicating potential influence of the predictor on changes in the outcome, were included into a multivariable PPO logistic regression model (to get ‘adjusted’ estimates). In step 3, those predictor variables not included in step 2 were then included, one at a time, and if they significantly shifted the estimates of any predictor variable already in the multivariable model, by shifting the result for proportional odds assumption or the 95% confidence intervals (CI) of the odds ratio completely, then they were retained in the final model, otherwise they were not included. A χ2 test was used to present descriptive summaries of; independent variables by survey year; completeness of each of the 4 health indicators in the RtHB by survey year; and expected completeness (all 4 indicators recorded in the RtHB) by each independent variable.