Antihypertensive Pharmacotherapy Gaps in Nigeria: A Predictive Machine Learning Analysis of Treatment Uptake Amid Macroeconomic Shock, Using NDHS 2023-24
Nosa-Ihaza, E. A., Edeh, E. C., Eze, D. N., Nosa-Ihaza, U. N.
health systems and quality improvement
Quick verdict
First NDHS 2023-24 hypertension cascade ML analysis: sound survey methods but no multiplicity control, missing baseline AUC, and overreached shock narrative.
Summary
This preprint analyzes treatment non-uptake among Nigerian adults diagnosed with hypertension, using the 2023-24 Nigeria Demographic and Health Survey (2,975 diagnosed women and 529 diagnosed men aged 15-49). Because that survey round recorded no blood-pressure measurements, the outcome is restricted to 'Gap 2': diagnosed individuals not currently taking antihypertensive medication. The authors fit survey-weighted logistic regressions separately by sex, formally test sex-by-predictor interactions in a pooled model, and compare elastic net, random forest, and XGBoost against weighted logistic regression using held-out AUC. Three findings stand out: non-uptake odds rose 14% per fieldwork month for women (OR=1.14, 95% CI 1.06-1.22) but not men (interaction p=0.016); geopolitical zones protective for women were risk factors for men (ORs 2.28-5.77; all zone-by-sex interactions p≤0.007); and diabetes was protective for women (OR=0.34). The ML algorithms converged at modest discrimination (women AUC 0.61-0.63; men 0.56-0.59), with no payoff from complexity. Key limitations: no multiple-comparison control across many tests; the logistic baseline's own AUC is never reported; men's AUC rests on a 104-person test set without confidence intervals; the fieldwork-timing 'natural experiment' rules out only geographic rollout; and self-reported data plus a cross-sectional design preclude causal claims. No code is shared; DHS data require registered access.
Review complete
Assessment at a glance
6
risk flags identified
● high 2
● medium 3
● low 1
5 strengths · 5 weaknesses · 6 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Odds of antihypertensive treatment non-uptake among diagnosed women rose 14% per fieldwork month during the subsidy-removal/naira-devaluation period (OR=1.14, 95% CI 1.06-1.22), with a formal sex-by-timing interaction (p=0.016) and no timing effect among men.
- Geopolitical zones protective against non-uptake for women (North East OR=0.60, North West OR=0.65) are significant risk factors for men (ORs 2.28-5.77), with all five zone-by-sex interactions significant (p≤0.007) — the largest associations in the study.
- Diabetes comorbidity is strongly protective against non-uptake for women (OR=0.34, p<0.001), with no confirmed sex interaction (p=0.170), supporting integrated opportunistic NCD screening.
- Elastic net, random forest, and XGBoost show statistically indistinguishable, modest discrimination (women AUC 0.61-0.63; men 0.56-0.59), so tree-based complexity adds no predictive value for this behavioral outcome, in contrast to the WHO STEPS CVD-risk literature.
Strengths
First national evidence from a brand-new data source, validated against published benchmarks
The paper is the first hypertension-cascade analysis of NDHS 2023-24; §4.2 independently reproduces the Final Report's benchmarks (7.97% weighted diagnosis prevalence vs ~8% published; 46.36% female non-treatment vs ~46%; 50.8% male treatment vs 51%), lending credibility to outcome construction from chd02/chd05 (women) and mchd02/mchd05 (men).
Formal interaction testing instead of the usual significance-comparison fallacy
§4.5 pools both sexes and tests sex-by-predictor interactions directly (e.g., timing-by-sex OR=0.82, p=0.016; all five zone interactions p≤0.007), and honestly reports the wealth interaction as NOT supported (OR=1.20, p=0.730) despite stratified significance in women only — a discipline rare in this literature.
Correct and transparent survey-design handling
Honest negative ML result and unstable-signal interpretation
Sensitivity specifications actually check robustness
Weaknesses
No multiplicity control across a large testing grid
ML evaluation is incomplete and internally inconsistent
The 'natural experiment' framing exceeds the design
Internal contradictions between abstract and results
Differential outcome ascertainment is a live bias threat, not just a mechanism
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This medRxiv preprint is, plausibly for the first time, a national-level analysis of the hypertension care cascade in the newly released Nigeria Demographic and Health Survey 2023-24. Because that round recorded no blood-pressure measurements, the outcome is restricted to 'Gap 2': the share of adults told by a health worker they have hypertension who are not currently taking antihypertensive medication (§3.3). Three claims structure the paper: (1) fieldwork timing — used as a real-time proxy for the fuel-subsidy-removal/naira-devaluation shock — raised non-uptake odds for women by 14% per month (OR=1.14, 95% CI 1.06-1.22), with no effect in men; (2) zones protective for women (North East OR=0.60; North West OR=0.65) are significant risk factors for men (ORs 2.28-5.77), confirmed by formal sex-by-zone interactions (all p≤0.007); (3) elastic net, random forest and XGBoost converge at modest AUC (0.61-0.63 women; 0.56-0.59 men), so algorithmic complexity adds nothing here — an explicit contrast with the WHO STEPS CVD-risk literature the authors cite (§5.3).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 6 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 6 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is 'Gap 2' in this study?
Gap 2 is the share of adults who were told by a doctor or health worker that they have hypertension but are not currently taking antihypertensive medication. This paper models it only among diagnosed respondents in Nigeria's 2023-24 DHS (2,975 women and 529 men aged 15-49); blood pressure itself was not measured in that survey round, so screening (Gap 1) and control (Gap 3) are out of scope.
Did the machine learning models beat logistic regression?
No. Elastic net, random forest and XGBoost achieved nearly identical AUCs on held-out data (0.61-0.63 for women; 0.56-0.59 for men, close to chance), so the added complexity bought no predictive gain. The authors attribute this to a mostly categorical sociodemographic predictor set and a behavioral outcome, in contrast to studies predicting composite clinical CVD risk.
Does the paper prove the economic shock caused treatment interruptions?
No. The odds of non-uptake among diagnosed women rose about 14% per calendar month of fieldwork (Dec 2023-May 2024), coinciding with fuel-subsidy removal and naira devaluation, and the effect survived a check for uneven geographic rollout. But this is an association from interview timing in a cross-sectional survey — the authors themselves state it rules out only one alternative explanation, not causal proof.
Is the NDHS data and analysis code publicly available?
The survey microdata are free but not open: access requires registration and approval from the DHS Program (dhsprogram.com). The authors used the women's (IR) and men's (MR) recode files. No analysis code or scripts were released, so exact re-execution is not currently possible from the paper alone.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1-3 (sample flow, forest plot, AUC comparison) appear only as captions in the extracted text; their visual content could not be verified, though all key values are restated in prose sections.
- Appendix materials and the country-specific questionnaire referenced in §3.3-3.4 were not part of the extracted text and could not be checked.
- Some editorial duplications in §3.6/§3.7 (e.g., the two XGBoost grids) could be extraction artifacts rather than manuscript errors; they are flagged as written.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.