Housing First Cost-Effectiveness Is Modelled With Offsets Added, Not Netted: One Lifetime Microsimulation Carries the Case

Question: How do cost-effectiveness analyses of Housing First and other permanent supportive housing interventions for homeless populations with substance use disorders model healthcare cost offsets and quality-adjusted life-years?

Answer

Across the retrieved papers, the accounting logic is not the one the question anticipates: the only cost-effectiveness analysis in this literature does not model offset savings at all, it adds both housing costs and healthcare costs to the intervention arm and asks what health gain is purchased per dollar. Ling et al. build a lifetime, individual-level microsimulation of the 18,928 people experiencing homelessness with opioid use disorder in California, count both housing and healthcare costs, and report an incremental cost-effectiveness ratio of $31,083 per quality-adjusted life-year gained, "well below conventional willingness-to-pay thresholds" [1]. The value claim therefore rests on cost-effectiveness, not on budget neutrality — the intervention is a good buy, not a self-financing one. QALYs in that model are carried overwhelmingly by the mortality channel rather than by measured morbidity: the intervention increases mean age at death by 0.34 years and averts 39 fatal overdoses in the cohort, and the population is characterised by "substantially lower quality-adjusted life expectancy" at baseline [1]. Everything else retrieved supplies either the mortality parameters such a model must be calibrated to or indirect, short-horizon evidence on utilisation that a credible offset assumption would have to draw on, and none of it reports a utility-based QALY for this population. Taken together, the papers support a narrower conclusion than the policy literature usually asserts: the modelled case for Housing First is a cost-effectiveness case built on overdose mortality averted, with the offset question left as a parameter assumption rather than an estimated quantity.

The offset is handled by inclusion, not by netting — and the programme is not shown to pay for itself

Ling et al. explicitly "accounted for both housing and healthcare costs," which means healthcare spending in the housed state is a cost to be added against the intervention, not a saving to be subtracted [1]. The reported ratio of $31,083 per QALY gained is the arithmetic consequence: the intervention buys health at a favourable price while consuming net resources [1]. The authors' conclusion is framed accordingly — that housing "is a high-value health intervention" and should be supported "as part of integrated responses" — a value framing, not a savings framing [1].

QALYs are generated by averting overdose deaths, so the mortality parameters do all the work

The health gain in the model is expressed as increased mean age at death (0.34 years) and 39 averted fatal overdoses, against a baseline of substantially lower quality-adjusted life expectancy among people experiencing homelessness [1]. The plausibility of that gain depends entirely on the mortality risk being modelled, and the meta-analytic evidence indicates the relevant relative risks are very large and highly cause-specific: all-cause mortality RR 2.12 in people exposed to homelessness, rising to 21.36 for deaths due to psychoactive substance use disorder, 10.80 for drug overdose, and 13.15 for accidental injuries, with the highest all-cause risk among people who had slept rough (RR 7.63) [5]. Because the elevated risk is concentrated in exactly the causes a housing intervention is plausibly able to interrupt, small differences in the assumed effect of housing on overdose risk translate into large differences in QALYs gained — the sensitivity of the $31,083 figure is therefore concentrated in a single parameter [1] [5].

The counterfactual is expanded access, and cohort size is set by a local ballot measure rather than by epidemiology

The modelled cohort is the 18,928 people experiencing homelessness with co-occurring opioid use disorder in California, derived from an estimated ten percent prevalence of OUD among that population, and the size of the housing increase is calibrated to Los Angeles County's Measure A goals [1]. The comparator is therefore incremental expansion of permanent housing and shelter access under a Housing First approach that does not require OUD treatment or abstinence, not a comparison of housing against no housing [1]. This makes the ICER a marginal estimate conditional on a specific policy target, and the authors' policy inference is scoped to California specifically [1].

Utilisation evidence that could anchor offsets is short-horizon and comes from adjacent interventions

The closest empirical analogue to an offset in the retrieved set is a street-medicine programme: after buprenorphine initiation, cumulative emergency department visits fell from 221 to 191 over twelve months, OUD-related visits from 79 to 44, and overdose-related visits from 26 to 13, with the OUD-related decline statistically significant [4]. That is a twelve-month, single-programme, pre-post comparison of one service component, not a housing intervention, and the reductions in total and overdose visits were not significant [4]. On the cost side, an integrated housing-plus-health programme for people with criminal-legal involvement was described as "time- and cost-intensive to deliver," with high-severity substance use predicting lower odds of being housed at follow-up (AOR 0.3) [3].

Service demand in supportive housing points toward offsets that are negative for part of the cost base

Residents of subsidised permanent supportive housing report strong interest in on-site dental and vision care (over 90%) and adult primary care accepting Medicaid (91.4%), in a population where most earn under $20,000 annually [2]. Under a model that counts healthcare costs in the housed state, this expressed demand is a cost of the intervention rather than a saving from it, and it sits in tension with any assumption that rehousing mechanically reduces healthcare spending [2] [1].

Conflicts and gaps

The central conflict is directional: the microsimulation's inclusive accounting yields a positive net cost per QALY [1], while the utilisation literature offers only narrow, service-specific reductions in acute contacts that are insufficient to establish whether total healthcare spending falls after rehousing [4]. No retrieved paper reports utility weights or EQ-5D-type measurements elicited in a homeless population with substance use disorders; the quality adjustment in the one CEA is inherited from quality-adjusted life expectancy rather than measured within the modelled cohort [1]. No retrieved paper compares Housing First against permanent supportive housing as distinct strategies, so the question's implicit contrast is untested. The parameter transfer is also temporally mismatched: offset-relevant evidence spans twelve-month follow-up [4] [3], whereas the cost-effectiveness model runs over a lifetime [1]. Finally, the model is confined to one state with a specific OUD prevalence assumption [1], and the mortality meta-analysis shows that risk varies sharply by shelter type — rough sleeping RR 7.63 versus hostels 3.44 — so a single population-average survival gain conceals heterogeneity that the housing literature itself documents as consequential for placement [5] [3].

What would change the answer

A staggered rollout of expanded housing access, analysed as a difference-in-differences or target-trial emulation against linked administrative hospital, emergency department, and Medicaid claims, would replace the modelled offset with an estimated one and determine its sign; the ballot-funded expansion used as the model's calibration target is the natural candidate [1]. Utility elicitation with a preference-based instrument administered longitudinally to housed and unhoused people with substance use disorders would allow quality-adjusted life expectancy to be measured rather than assumed [1]. Mortality-linkage follow-up of a housed cohort would directly test the 0.34-year survival gain and the overdose-aversion mechanism, using the cause-specific relative risks as the calibration benchmark [1] [5]. Probabilistic sensitivity analysis with value-of-information on the offset and overdose-risk parameters would establish how much of the $31,083 per QALY conclusion survives uncertainty in exactly the two inputs that drive it [1]. And a model with take-up stratified by substance use severity would test whether the people with the largest potential mortality benefit are the least likely to be housed [3].

References

[1] Ling, Y., Tsung, S., Alarid-Escudero, F., Goldhaber-Fiebert, J. D., Brandeau, M. L., Reitsma, M. B.. "Cost-Effectiveness of Expanding Housing Access for People Experiencing Homelessness and Opioid Use Disorder in California". medRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.10.15.26365505

[2] Clouse, K., Kimpel, C., Chavez, L., Terrell, T., Landivar Donato, J., Hasbrooke, N., Frenia, M., Elrod, B., Ketel, C.. "Demographic and economic characteristics of residents of four subsidized housing facilities in Nashville, TN, and their preferences for co-located services". medRxiv. 2025-08-26T00:00:00.000Z. URL: https://doi.org/10.1101/2025.08.23.25334285

[3] Fan, A. Y., Flax, C., Ibrahim, N., Tracey, D., Hernandez, A., Moscariello, S., Price, C. R., Meyer, J. P.. "Effect of an integrated housing intervention for people involved in the criminal-legal system who have housing instability". medRxiv. 2026-02-27T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.25.26347012

[4] Wallace, C. R., King, J., Nichols, T., Abrams, E. A., Shaw, S., Hoeflinger, N., Mullangi, V., Sandoval, B., Matzke, J., Kauffman, E., Mynatt, I., Pelc, M., Cantzler, R.. "Buprenorphine Initiated via Street Medicine Associated with Reduced Opioid-Related Morbidity Among People Experiencing Homelessness". medRxiv. 2025-09-19T00:00:00.000Z. URL: https://doi.org/10.1101/2025.09.17.25336012

[5] White, J., Moriarty, Y., Lau, M., Weightman, A., Kiseleva, M., Cannings-John, R., Palmer, A., Batty, G. D.. "Homelessness, type of homelessness, and risk of cause-specific mortality: a systematic review and meta-analysis of 116 studies comprising 2,563,633 homeless people and 129,292,553 population controls". medRxiv. 2025-04-22T00:00:00.000Z. URL: https://doi.org/10.1101/2025.04.22.25326193