The 1930s Map Still Sorts Who Dies of an Overdose in Massachusetts

The federal government's Home Owners' Loan Corporation drew its residential security maps in the late 1930s, coloring neighborhoods green, blue, yellow or red according to how safe it considered mortgage lending there. Red meant hazardous. The agency's assessors wrote the reasons in the margins: older housing, immigrant families, Black residents.

Eighty-odd years later, a team of Massachusetts researchers took those grades, carried them forward onto 2010 census tracts, and compared them with every opioid-related overdose death recorded in the state between 2015 and 2019. The ordering held.

Compared with tracts the HOLC graded Best, overdose death rates were 53 percent higher in tracts graded Still Desirable, 61 percent higher in those graded Declining, and 86 percent higher in those graded Hazardous, after adjusting for demographics and for each tract's social vulnerability. The study appears in the Journal of Health Care for the Poor and Underserved, and was publicized by News-Medical on Nov. 4.

The finding is not that a 1930s map causes overdoses. It is that the map turned out to be a durable instrument for predicting where they would happen — which is a different and more uncomfortable claim.

Turning a paper map into a 2026 number

The mechanics matter, because this is the kind of study that can be made to say anything if the method is sloppy.

The authors assigned a redlining grade to every 2010 census tract in Massachusetts using a weighting method, since tract boundaries and HOLC polygons do not line up and a single tract can straddle two grades. They linked those grades to decedent records for 2015 through 2019 and to census tract population estimates, then ran negative binomial regression — the standard choice for count data that is overdispersed, which overdose deaths are.

Two technical decisions raise the paper's credibility. First, they adjusted for census tract social vulnerability, which means the result is not simply a restatement of "poorer neighborhoods have more overdoses." The 86 percent gap survived that adjustment. Second, they used robust standard errors to account for spatial autocorrelation, having confirmed it was present with a Moran's I test significant below .001. Neighboring tracts resemble each other, and a model that ignores this will manufacture confidence it has not earned. This one does not.

HOLC grade	Adjusted incidence rate ratio	95 percent confidence interval
Best	reference	—
Still Desirable	1.53	1.03–2.28
Declining	1.61	1.07–2.43
Hazardous	1.86	1.18–2.93

What the paper does not report in its abstract is the absolute rate in each grade. That gap matters, because a relative ratio of 1.86 on a small base is a different public health problem than the same ratio on a large one, and readers cannot compute the difference from what has been published so far.

What the authors controlled for, and what that costs

Social vulnerability indices bundle together income, housing quality, vehicle access, crowding and similar measures. Adjusting for one is the right call, and it also removes some of the pathway the paper is trying to measure. Disinvestment is the mechanism by which redlining is supposed to work; a covariate that captures the downstream poverty produced by that disinvestment will absorb part of the effect.

The association that survives is therefore best read as a floor, not a ceiling. The true gap between a Hazardous and a Best tract is plausibly larger than 86 percent, not smaller — because the study stripped out a channel through which the original policy operates.

The 2010 census tract problem

Assigning a 1930s grade to a 2010 geography requires a modelling choice, and the paper's weighting method is one of several defensible approaches. Tracts that were entirely built after the maps were drawn, or that were split and merged across eight decades, carry more uncertainty than the point estimates convey. That is a limit of the whole redlining literature rather than of this paper, but it is the reason a single study should not be used to rank neighborhoods or to target individuals.

Ninety years is not a metaphor

The pathways connecting a mortgage-lending classification to a fatal overdose are concrete and mostly boring, which is why they are easy to ignore.

Housing policy determined where Black and immigrant families could buy, and therefore where wealth accumulated or did not. It determined school funding, because American schools are funded by local property taxes. It shaped where pharmacies opened and closed, and where physicians chose to practice. It shaped policing, and therefore the criminal records that follow people into employment, housing and, in many states, eligibility for public benefits. It shaped whether a methadone clinic that wanted to open in the 1970s was welcomed or run out of town, which is a live history in several Northeast cities.

Layer a synthetic opioid supply on top of that geography and the death rates sort themselves. Not because the map is magic, but because the map is a compressed record of who got to accumulate safety.

Chicago found the same shape with a different instrument

This is not an isolated result. In November 2025, a Johns Hopkins-led team published an analysis in JAMA Health Forum of 796 Chicago census tracts, combining historical redlining with contemporary racialized economic segregation into a single index across an eighty-year span.

Their finding had a similar geometry. Tracts that ranked high on structural racism in both the past and the present averaged 2.6 more fatal overdoses per tract than consistently advantaged tracts in 2017 to 2019, an average marginal effect of 2.60 with a confidence interval of 2.02 to 3.19. During the pandemic years of 2020 to 2022, the group that fared worst had changed: tracts that had been advantaged historically but experienced high present-day segregation averaged 3.81 more fatal overdoses per tract. Every group's burden rose during COVID-19.

Read the two papers together and a useful distinction appears. Massachusetts measured whether an old grade still sorts outcomes. Chicago measured what happens when an old grade and a current one disagree — and found that present-day segregation alone was enough to move a neighborhood into the worst overdose category even when its historical grade was favorable. Redlining is not destiny. It is one load-bearing beam in a structure that is still being built.

The limits, stated plainly

This is an ecological study. Exposure is measured at the census tract and so is the outcome. It cannot show that any individual died because of a lending decision made before their grandparents were born, and it cannot rule out that both the grade and the deaths are produced by something the model did not measure.

The window is also a problem in both directions. Data end in 2019, before COVID-19 disrupted drug supply and treatment access, and before fentanyl and counterfeit pills reshaped the market. If the association was this strong using deaths from a period with lower overall mortality, the current version of the same gradient is likely steeper — an inference the paper cannot test and the next one should.

There is also a selection question the authors do not address in the abstract: Massachusetts is a comparatively high-treatment, high-insurance state with an unusually dense addiction medicine infrastructure. A gradient that survives there might be even sharper in states with fewer providers, or it might be masked where everyone's access is equally poor.

Why this lands on treatment policy

Overdose prevention money is, at this point, mostly geographic money. Opioid settlement funds are distributed to counties and municipalities, medication-assisted treatment is delivered at fixed addresses, and mobile units follow routes someone chose. The argument this paper supports is not that individual tracts should be ranked, but that siting decisions made on current population counts alone will reproduce a pattern that is already ninety years old.

For the clinicians and program directors who will read it, the practical version is narrower. If a county is deciding where to place a bridge clinic, an opioid addiction treatment program or a medication-assisted treatment prescriber, the historical grade of the surrounding tracts is a variable they already have access to, costs nothing to obtain, and correlates with overdose mortality more tightly than most of what goes into a needs assessment.

The uncomfortable part is what the finding implies about how long policy lasts. A grading system designed to last a generation is still sorting death certificates eight decades later, and the interventions that could interrupt it — housing, pharmacy access, clinic siting, prescriber distribution — operate on a timeline of years. The map outlives the people who drew it. The question the study leaves open is whether anything short of redrawing the underlying geography will change what it predicts.