Beyond disease counts: multimorbidity patterns and self-rated health in South African adults-a network analysis of the 2016 demographic and health survey.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
BACKGROUND: Multimorbidity is rising rapidly in sub-Saharan Africa, where infectious and non-communicable disease burdens overlap. Whether the pattern of co-occurring conditions carries information beyond a simple disease count is unclear. OBJECTIVE: To map patterns of co-occurring chronic conditions in South African adults, test for additive interactions between condition pairs, and compare network-derived measures with the disease count as correlates of poor self-rated health. METHODS: Cross-sectional analysis of the nationally representative 2016 South Africa Demographic and Health Survey. The analytic sample comprised 5,650 adults aged 18 years and older with complete data on 12 chronic conditions ascertained by self-report and biomarker measurement. The outcome was self-rated health, dichotomised as poor vs. average, good or excellent. Disease networks were constructed from phi coefficients with Louvain community detection; cluster stability used 1,000 design-consistent bootstrap replicates. Survey-weighted logistic regression adjusted for age, sex and residence compared the disease count with network-based measures (per 1-SD increase), using the area under the receiver operating characteristic curve (AUC), DeLong tests and cross-validation. RESULTS: Three disease clusters emerged: infectious (tuberculosis, HIV, anemia), cardiometabolic and pain (hypertension, diabetes, chronic pain), and cardiovascular-respiratory-neoplastic (stroke, heart attack, high cholesterol, chronic bronchitis, asthma, cancer). The infectious cluster co-clustered in 100% of bootstrap replicates, and network structure differed by sex (permutation p = 0.001) and residence (p = 0.031). Adjusted odds ratios for poor self-rated health were 1.66 (95% CI 1.41-1.95) per SD of disease count, 1.55 (1.31-1.82) per SD of network score and 1.24 (1.03-1.49) per SD of max centrality; the network measures were highly correlated with disease count (r = 0.96) and did not improve discrimination (AUC 0.756 vs. 0.758; DeLong p = 0.26), a finding confirmed by cross-validation. One additive interaction remained after false-discovery-rate correction: stroke with anemia (+0.55 on the risk-difference scale, 95% CI 0.24-0.86). CONCLUSIONS: Network analysis revealed robust disease clustering, most notably a stable infectious cluster, and subgroup-specific patterns relevant to integrated care planning. Individual-level network measures did not outperform a simple disease count. Disease counts remain a pragmatic risk marker; pattern-based measures merit prospective evaluation before clinical adoption.