Co-Occurrence by Construction: Disability and Mental Illness in LMIC Household Surveys

Question: How do national household surveys in low- and middle-income countries measure disability and mental illness co-occurrence, and what methodological concerns arise from using self-reported single-item measures?

Answer

Taken together, the retrieved papers show that LMIC household surveys do not really measure disability–mental illness co-occurrence as a clinical phenomenon; they cross-tabulate two coarse self-reported flags produced by instrument modules designed for other purposes. The dominant architecture is a functioning module — most often the Washington Group Short Set — administered to a questionnaire subsample, paired with a brief self-reported mental/psychological illness item, and only one retrieved study (Somaliland 2020) actually operationalises co-occurrence from such data, doing so by comparing observed joint prevalence against statistical independence [2]. Across these papers the methodological concerns cluster into four: the disability module may itself contain a mental domain, so the two constructs are not definitionally independent [2]; estimates swing widely with threshold and definition choices [11] [5]; self-report and validated screening instruments applied to the same population diverge substantially [6]; and symptom-report items are not measurement-equivalent across socioeconomic and ethnic groups [10], with survey design itself excluding some disabled respondents [9]. The response emerging in this literature is not a better single item but hybrid self-report-plus-clinical assessment [8].

The standard LMIC instrument architecture pairs a functioning module with one self-reported illness item

Disability in these surveys is operationalised as a set of functional domains with a binary threshold. In the Kenya DHS 2022 person recode, disability was measured with the Washington Group Short Set and 6.7% of 40,197 adults met the WG threshold, with severity categories contrasted directly [1]. The Malawi DHS 2024 assessed maternal functional difficulty on the six WG-SS domains of seeing, hearing, walking, cognition, self-care and communication, classified as severe versus no difficulty, yielding a weighted prevalence of 3.74% among 4,361 mother–child dyads [5]. The Somaliland 2020 survey used a seven-domain module — sight, hearing, speech, learning, mobility, self-care and mental — defining reported disability as at least one positive response, giving a weighted prevalence of 7.30% (95% CI 6.65–8.02) that rose from 4.15% at ages 18–29 to 23.96% at ages 60 and above [2]. In that survey, reported mental/psychological illness was carried as a separate measure with a weighted prevalence of 0.71% (95% CI 0.50–1.00) [2].

In the only retrieved co-occurrence analysis, the two constructs are not independent by construction

Co-occurrence in the Somaliland study was evaluated by comparing observed joint prevalence with that expected under statistical independence and by survey-weighted logistic regression [2]. The same survey's disability definition includes a mental domain while mental/psychological illness is analysed as a distinct measure, and the report does not state whether the underlying items are separate [2]. This matters because disability measurement in these instruments is explicitly functioning-based rather than condition-based elsewhere — the FNAT was developed precisely because reliable population-level data on functional difficulties and service need are lacking, and it treats mental health as one of seven domains alongside vision, hearing, mobility, communication, cognition and self-care [8]. Treating a mental-health domain inside a disability index and a mental-illness item outside it as two variables to be tested for association assumes exactly the separation that the instrument design does not guarantee.

Estimates are threshold- and definition-dependent, and self-report diverges from validated screening

Applying different disability definitions to the same construct produces sharply different prevalence. Among 104 women living with HIV assessed with the Episodic Disability Questionnaire, disability prevalence was 79.81% under a moderate WHODAS 2.0 threshold but 41.75% under the severe Equality Act definition, from the same respondents [11]. Where self-report and clinically validated screening are administered side by side, the two do not agree: in a purpose-collected UK claimant survey, 72.6% said they were currently affected by a mental health condition while 61.2% screened positive for moderate/severe depression or anxiety on GAD-7/PHQ-9, both roughly twice general-population levels [6]. The direction of that divergence runs opposite to what very low self-reported mental illness prevalence in a low-income setting would imply — South African household survey data place depression point prevalence at 5.1% in 2002 and 4.5% in 2024 [12] — and no retrieved paper reconciles these magnitudes at the item level.

Single-item symptom reports are not measurement-invariant, and survey design excludes respondents

Symptom-report items can function differently across groups at equivalent underlying risk: in the K-SADS screening interview, nearly 50% of items functioned differently across Hispanic, non-Hispanic white and non-Hispanic Black youths despite a scalar-equivalent one-factor structure, and socioeconomic factors explained many of those measurement differences [10]. Survey administration itself introduces exclusion: analysis of HINTS 2024 concluded that true disability representation is likely not possible because of inaccessible survey design, even while documenting that people with multiple disabilities, chronic pain or deafness had higher odds of frustration and difficulty understanding health information [9]. Self-reported contact measures also do not carry clinical meaning: in the 2022 Peruvian National Mental Health Survey, lifetime and 12-month contact gaps were 61.0% and 84.5%, only 37.3% of those meeting criteria for a 12-month disorder reported self-perceived need, and the authors state explicitly that these contact measures do not establish adequate, continuous or effective treatment [7].

The proposed remedy is hybrid self-report plus clinical assessment, not a refined single item

The Functional Needs Assessment Tool was built specifically to close the data gap on population-level functional difficulties and need for services and assistive products, and it combines self-reported and clinical assessment tools across seven domains including mental health, using a two-stage population-based assessment collected through a tablet application [8]. Its developers caution that the tool still needs pilot testing to establish feasibility [8]. Data-integrative modelling has been proposed as a complementary route: the South African depression analysis calibrated a system dynamics model to nine nationally representative household surveys (2002–2024) precisely because prevalence and treatment data were inconsistent and fragmentary, estimating a decline from 5.1% to 4.5% and a lifetime prevalence of 70.6% [12]. At the metric layer, the DALY — the standard vehicle for combining mental illness and disability burden — has itself been subjected to a methodological critique questioning its robustness and its role in upholding structural ableism [3].

Conflicts and gaps

The retrieved papers disagree on the direction of self-report bias. One side-by-side comparison finds self-reported mental health conditions more prevalent than screening-tool caseness (72.6% vs 61.2%) [6]; the Somaliland data report mental/psychological illness at 0.71% of adults [2], while South African household surveys support depression prevalence an order of magnitude higher [12]. No retrieved study validates a brief mental-illness item against a diagnostic interview in an LMIC household survey, so the source of this divergence cannot be attributed. Second, co-occurrence methodology rests on a single country study using an observed-versus-expected comparison [2]; nothing retrieved replicates it. Third, the disability evidence base is geographically skewed: a systematic review of disability and maternal and child mortality found that almost all data came from nationally representative sources in high-income countries and flagged the lack of evidence particularly from LMICs [4]. Fourth, disability is often fielded only in a long-questionnaire subsample [1], which constrains any co-occurrence cell that requires both a disability and a mental-health measure.

What would change the answer

A criterion-validation study embedded in an existing DHS or similar long-questionnaire subsample — administering a structured diagnostic interview alongside the WG-SS and the single mental-illness item in the same respondents — would establish the sensitivity and specificity of the current items and therefore whether reported co-occurrence reflects morbidity or item behaviour, following the two-stage self-report-then-clinical design already specified for the FNAT [8]. Formal measurement-invariance testing of both the disability module and the mental-illness item across wealth, education, residence and ethnicity — the approach applied to K-SADS items [10] — would show whether cross-group comparisons of co-occurrence are admissible at all. Pooled multi-country analysis using latent variable models rather than observed-versus-expected joint prevalence would separate the functioning and mental-health constructs empirically instead of assuming them [2]. Finally, calibration against recorded treatment data, as done for South African depression, would benchmark survey-based prevalence against service-system evidence and indicate the direction of any reporting bias [12].

References

[1] Amollo, N. W., Ouma, J. O.. "Insurance coverage, healthcare use, and outpatient payment among adults with disability in Kenya before the SHA transition: a secondary analysis of KDHS 2022". medRxiv. 2026-09-21T00:00:00.000Z. URL: https://doi.org/10.64898/2026.09.20.26363504

[2] Saed, A. A., Osman, O. A., Adan, A. M., Suleiman, Y. I., Mohamed, A. I.. "Adult disability and mental/psychological illness in Somaliland socioeconomic inequalities and co-occurrence in the 2020 Somaliland Health and Demographic Survey". medRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.09.24.26363961

[3] Dielle Lundberg. "A Methodological Critique of the Disability-Adjusted Life Year (DALY): Robust Science or a Symptom of Structural Ableism in Public Health?". socarxiv. 2025-07-31T08:49:24.374Z. URL: https://osf.io/nzk62_v1

[4] Rotenberg, S., Chilufya-Moyo, M., Valentine, A., Smythe, T., Forde, I., Mitra, M., Kuper, H.. "Maternal mortality and adverse child outcomes for women with disabilities: A systematic review and meta-analysis". medRxiv. 2026-08-27T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.25.26361292

[5] Samu, G. C., Karki, R., Ng'ambi, N. C., Abiona, M. M., Karki, S., Samu, S., Karki, B. K., Mphemvu, G.. "Maternal Functional Difficulty and Child Undernutrition in Malawi:A Nationally Representative Cross-Sectional Study( 2024 DHS)". medRxiv. 2026-08-05T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.03.26359624

[6] Ben Geiger, Lisa Scullion, Daniel Edmiston, Robert de Vries, K Summers, Jo Ingold, David Young. "Mental ill-health among welfare claimants in the UK: The extent, nature, and impact on claimant experiences". socarxiv. 2024-12-13T10:59:59.301Z. URL: https://osf.io/8qygm_v1

[7] Ruiz-Grosso, P., Macedo-Orrego, L., Rodriguez-Vargas, D., Rivera-Encinas, M., Arosemena, A., Carazas-Vera, M., Sagastegui, A., Zevallos-Bustamante, S.. "Mental health contact gaps among adults in Peru: a cross-sectional analysis of a nationally representative survey". medRxiv. 2026-08-22T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.19.26360808

[8] Boggs, D., Birabwa, A., Adkins, S., Atijosan-Ayodele, O., Bulathwela, S., de Cates, C., Foster, A., Kuper, H., Holloway, C., Mugisha, J., Polack, S.. "Development of a Functional Needs Assessment Tool to estimate population level functional difficulties and need for services and assistive products". medRxiv. 2026-07-17T00:00:00.000Z. URL: https://doi.org/10.64898/2026.07.10.26357317

[9] Velat, M., James, T. G.. "Health Information Seeking Experiences Among People with Disabilities: Results from the HINTS 2024". medRxiv. 2025-10-15T00:00:00.000Z. URL: https://doi.org/10.1101/2025.10.14.25338014

[10] Quanfa He, James Li. "K-SADS in Context: Testing Measurement Equivalence Across Race, Ethnicity, and Socioeconomic Factors". psyarxiv. 2025-08-28T17:57:12.098Z. URL: https://doi.org/10.1037/pas0001514

[11] Brown, D. A., Tariq, S., Boffito, M., Asboe, D., Milinkovic, A., Nwokolo, N., Flavell, C., Strachan, S., Avery, L., O'Brien, K. K., Harding, R.. "Evaluation of the psychometric properties of the Episodic Disability Questionnaire (EDQ) among women living with HIV in the United Kingdom: a self-reported repeated measure study". medRxiv. 2025-11-02T00:00:00.000Z. URL: https://doi.org/10.1101/2025.10.30.25339187

[12] Johnson, L. F., Giovenco, D., Eyal, K., Craig, A., Petersen, I., Tlali, M., Levitt, N. S., Bachmann, M., Haas, A. D., Fairall, L.. "Patterns of depression prevalence and antidepressant use in South Africa, 2002-2024: a system dynamics modelling perspective". medRxiv. 2026-03-09T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.08.26347878