Diagnosis as the Treatment: Threshold Evidence on Chronic-Disease Labels and Labor Supply

Question: What do quasi-experimental studies using biomarker-based diagnostic thresholds find about the labor supply and employment effects of receiving a chronic disease diagnosis, such as hypertension or hyperlipidemia?

Answer

Taken together, the retrieved papers support a narrow and conditional answer. Where a biomarker threshold generates a diagnosis, labor supply appears to fall in the short run — but the only quasi-experimental estimate in this corpus is for diabetes, where a fuzzy regression discontinuity at the diagnostic threshold reduces working for pay by 22 percentage points and labor-force participation by 20.6 percentage points at two years, with no offsetting rise in retirement and no sustained improvement in cardiometabolic outcomes [2]. No retrieved study applies an equivalent threshold design to hypertension or hyperlipidemia, and the nearest hypertension evidence is associational and far smaller in magnitude [8]. Meanwhile, the running variables these designs rely on — a blood pressure or HbA1c reading against a cutpoint — are demonstrably sensitive to measurement and analytic choices at a scale that rivals the effects being estimated [3] [6]. The position of this survey is therefore that diagnosis functions as a distinct economic shock separable from underlying disease severity, that the existing evidence locates this shock primarily in behavior rather than in physical work capacity, and that its magnitude for hypertension and hyperlipidemia remains genuinely unmeasured rather than merely uncertain.

The one threshold-based labor supply estimate is large, negative, and transient

The diabetes study is the only retrieved paper that identifies labor supply off a biomarker cutpoint: it combines a fuzzy regression discontinuity in American Family Cohort electronic health records with Health and Retirement Survey panel data, finding the diagnosis discontinuity peaks at 13 percentage points within six months and becomes insignificant by 42 months [2]. The same paper reports that the labor supply reductions of 22 percentage points in working for pay and 20.6 percentage points in labor-force participation at the two-year follow-up survive several sensitivity checks but do not persist beyond two years [2]. Its authors explicitly frame the implication for policy evaluation: health screening for chronic conditions may temporarily reduce labor supply, which makes employment responses relevant to the broader costs and benefits of screening [2].

The work reduction is not accompanied by a health improvement, pointing to a labelling channel

The same study finds only specification- and treatment-window-dependent improvements in weight, BMI and serum glucose, and no pattern of sustained improvement in cardiometabolic outcomes following diagnosis and related care [2]. It also finds no corresponding change in retirement, which its authors read as respondents reducing work without retiring [2]. A labelling interpretation is reinforced by evidence that people with diabetes who were unaware of their condition before a study were less likely to report problems with mobility, self-care, pain or discomfort, and usual activities than those already diagnosed [10]. If becoming a patient changes reported functioning more than measured physiology, the labor supply response at a threshold is plausibly a response to the label and its attendant role expectations rather than to a change in work capacity.

Hypertension evidence is associational, and it does not reproduce the diabetes magnitude

In British birth cohort data, a longstanding illness reported at age 42 is associated with a 6 percentage-point increase in health-related economic inactivity a decade later, and mental ill-health with a 4.5 percentage-point increase, with magnitudes similar across the 1958 and 1970 cohorts except for high blood pressure [8]. That 6 percentage-point associational estimate is roughly a third of the 22 percentage-point discontinuity estimated for diabetes, and it bundles together conditions of very different symptom burden; high blood pressure is singled out as the exception in cross-cohort comparison [8]. The population exposed to this margin is nonetheless expanding, with diagnosed hypertension rising from 12.3% to 19.3% of Mexican adults between 2000 and 2023 while undiagnosed hypertension fell from 20.7% to 11% [4].

The threshold itself is a noisy treatment assignment, which biases these designs toward nulls

An executable multiverse analysis of India's National Family Health Survey shows that holding case definition, population and estimand fixed, five defensible blood-pressure reading-selection and quality-control algorithms move prevalence by 1.58 to 3.07 percentage points, which is 4.5 to 19.6 times the design-based standard error, and that only 70 of 480 sibling estimates fall inside a given specification's own 95% confidence interval [3]. Measurement modality shifts threshold crossing as well: matched primary-care patients measured with manual devices had a mean systolic pressure of 125.9 mmHg against 131.7 mmHg with automated devices, and 38.8% versus 33.2% fell below the 130/80 cutpoint [6]. Because such misclassification is not classical around a cutpoint, any hypertension or lipid discontinuity design inherits an attenuated first stage and a diluted reduced form [3] [6]. The aware–unaware margin is also large and poorly predicted by routine features, with 20.5% of a nationally representative US sample meeting criteria for probable undiagnosed hypertension and the best classifier reaching an AUC of only 0.611 [1].

Treatment effects run in the opposite direction, so the sign of the net effect is an empirical question, not a prior

Where the retrieved papers estimate the effect of treatment rather than diagnosis, the sign reverses: elective inpatient treatment is associated with sustained gains in monthly earnings and employment, reaching £1,805 more per month at 60 months in clinical haematology and a 29.9 percentage-point employment advantage in clinical oncology [7]. Longer waits for elective musculoskeletal treatment reduce annual employee earnings with no detectable effect on employment status or benefit receipt, implying that the adjustment margin there is hours or productivity rather than exit [9]. A diagnosis that triggers effective treatment and a diagnosis that merely relabels a person can therefore push labor supply in opposite directions, and the diabetes threshold result — work down, health flat — locates the observed effect on the labelling side [2].

Conflicts and gaps

The central conflict is one of magnitude and design. A 22 percentage-point drop in working for pay from a diagnostic threshold implies an employment response far larger than the 6 percentage-point associational gradient for longstanding illness in cohort data [2] [8], and no retrieved paper reconciles the two. The unaware-versus-diagnosed quality-of-life contrast is cross-sectional and could reflect reverse causation, since symptomatic people are the ones who present for testing [10]. Who gets diagnosed is itself selected: in Argentine national survey data, higher income quintiles had 22% greater odds of remaining undiagnosed at elevated glucose [5], which limits how far threshold-local estimates generalize. Nothing in the corpus estimates hours, earnings, or sickness absence at a blood-pressure or lipid cutpoint; hyperlipidemia is entirely absent despite its threshold-driven statin pathway; and the persistence question is confounded because the diagnosis discontinuity itself decays to insignificance by 42 months, so a "transient" labor effect and an evaporating treatment both load on the same short window [2].

What would change the answer

Three designs would settle the open questions. First, port the fuzzy discontinuity to hypertension and to LDL cholesterol, using the 2017 ACC/AHA 130/80 mmHg and 2013 cholesterol guideline cutpoints as running variables in EHR data linked to earnings — ideally exploiting the guideline change itself as a difference-in-discontinuity, since the same reading sits on one side of the old and new thresholds. Second, estimate those designs on repeated or automated readings so that the running variable is measured with error small relative to the threshold gap, given how much prevalence moves with reading-selection rules [3] [6]. Third, link diagnosis thresholds to administrative earnings, sick-leave and disability-benefit registers of the kind already used for elective treatment and waiting times [7] [9], which would distinguish exit from hours reduction and separate the diagnosis shock from the treatment response that follows it. Replication of the diabetes estimate in a second population and age band is the minimum condition for treating its 22 percentage points as a transportable parameter.

References

[1] Thakur, M., Aacharya, S., Tiwari, P., Sah, R., Yadav, P., Yadav, D., Thakur, B.. "Screening for Probable Undiagnosed Hypertension in US Adults Using Interpretable Machine Learning: An NHANES 2017-2018 Study". medRxiv. 2026-06-25T00:00:00.000Z. URL: https://doi.org/10.64898/2026.06.23.26356178

[2] Heitzig, C., Rehkopf, D.. "Diabetes and the Life-Course: Evidence from Panel Data and Electronic Health Records". medRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.06.06.26355069

[3] Dharit Shah. "Blood-pressure reading and quality-control choices move India's raised-blood-pressure prevalence several times beyond its sampling error: an executable multiverse analysis of NFHS-4 and NFHS-5". socarxiv. 2026-08-24T00:20:21.258Z. URL: https://osf.io/3g6n2_v1

[4] Fermin-Martinez, C. A., Nunez-Luna, A., C. Guerra, E., Ramirez-Garcia, D., Perezalonso-Espinosa, J., Zarco-Morales, K. P., Leon-Alvarez, M., Ponce-Acosta, C., De la Maza-Bustindui, N. S., Vargas-Vazquez, A., Antonio-Villa, N. E., Bello-Chavolla, O. Y.. "Evolving patterns of prevalence and management of hypertension phenotypes in Mexico: A two-decade analysis of nationally representative surveys". medRxiv. 2025-08-06T00:00:00.000Z. URL: https://doi.org/10.1101/2025.08.04.25333004

[5] Munoz Nigro, M. A.. "Undiagnosed Dysglycemia and Socioeconomic Status in Argentina: A Paradoxical Gradient in the 2018 National Survey of Risk Factors". medRxiv. 2026-03-10T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.03.26347343

[6] Lopez-Silva, C., Surapaneni, A., Shin, J.-I., Horwitz, L., Blecker, S., Flaherty, C., Foti, K., Grams, M. E., Chang, A. R.. "Underestimation of Blood Pressure and Stroke Risk by Manual Blood Pressure Measurement". medRxiv. 2026-02-26T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.24.26346929

[7] Aston, H. E., Bunk, H., Henstock, T., Madhukar, A., Batrakova, S., Memon, A., Nafilyan, V., Ayoubkhani, D.. "Impact of elective inpatient treatments on monthly earnings and employment: a national linked data study in England". medRxiv. 2025-11-25T00:00:00.000Z. URL: https://doi.org/10.1101/2025.11.19.25340566

[8] Gimeno, L., Bridger Staatz, C., Goisis, A., Dowd, J., Ploubidis, G. B.. "Chronic health conditions and health-related economic inactivity in midlife: Evidence from the 1958 and 1970 British birth cohorts". medRxiv. 2025-06-20T00:00:00.000Z. URL: https://doi.org/10.1101/2025.06.20.25329982

[9] Aston, H., Nafilyan, V., Meacock, R., Ayoubkhani, D.. "The impact of hospital waiting lists on labour market outcomes: quasi-experimental evidence from musculoskeletal treatment in England". medRxiv. 2026-10-07T00:00:00.000Z. URL: https://doi.org/10.64898/2026.10.05.26364590

[10] Sabin, L., Kumar Shaha, S., Kiadaliri, A., Kuddus, A., King, C., Zhou, X., Ahmed, N., Jennings, H., Morrison, J., Akter, K., Nahar, T., Azad, K., Fottrell, E., Haghparast-Bidgoli, H.. "Quality of life and its determinants among individuals with type 2 diabetes mellitus in rural Bangladesh". medRxiv. 2025-06-12T00:00:00.000Z. URL: https://doi.org/10.1101/2025.06.11.25329444