Chapter 7.3 · Where illness keeps company

Multiple chronic conditions

Weighted by adults, census tracts in the most deprived tenth were high on 5.7 of nine modeled chronic-condition prevalences on average in 2022, against 0.5 in the least deprived tenth, a count for tracts and not people.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 9 October 2026

Of the 82,781 census tracts in the data for 2022, 892 were in the highest 25% of tracts on the modeled prevalence among adults of each of nine chronic conditions. The nine are arthritis, cancer, current asthma, coronary heart disease, COPD, depression, diagnosed diabetes, obesity and stroke. If the conditions were unrelated across tracts, with each condition's observed share of high tracts, they would give 0.3 tracts high on all nine. That benchmark is a reference point, and real tracts are not independent draws, in part because the model fits every condition to the same inputs. The count describes places. A tract can be high on several conditions while no resident has more than one, so the count says nothing about how many conditions any one person has.

The count rises at every step of the deprivation index

The chapter ranks tracts on the Area Deprivation Index (ADI), a national percentile built from American Community Survey (ACS) data. Weighted by adults, a tract in the most deprived tenth of tracts is high on 5.7 of the nine conditions on average, against 0.5 in the least deprived tenth. The mean count rises at every step from the least deprived tenth to the most deprived, with the two largest steps at the deprived end.

Part of this gradient is mechanical. The ADI is built from ACS data, and the PLACES model applies survey responses to each tract's census population using inputs from the same census sources. The conditions and the index therefore move together partly by design. The gradient of each condition taken one at a time belongs to Chapter 7.1, Health behaviors and the supply of care. The count here adds only that the conditions which rise are high in the same tracts.

Figure 1

Weighted by adults, a tract in the most deprived tenth is high on 5.7 of nine conditions on average, against 0.5 in the least deprived tenth

Population-weighted mean number of the nine conditions that are high, by tenth of tracts on the ADI national percentile, 82,781 tracts, 2022

Mean number of high conditions (of nine)

Show data

Data for Weighted by adults, a tract in the most deprived tenth is high on 5.7 of nine conditions on average, against 0.5 in the least deprived tenth, Mean number of high conditions
conf_high	conf_low	decile	estimate	measure	n_units
0.550	0.380	1	0.450	count	8,274
0.700	0.550	2	0.620	count	8,118
1	0.780	3	0.880	count	8,511
1.10	0.920	4	1.01	count	8,261
1.39	1.19	5	1.29	count	8,189
1.83	1.60	6	1.71	count	8,318
2.46	2.22	7	2.34	count	8,305
3.34	3.07	8	3.21	count	8,238
4.49	4.21	9	4.35	count	8,354
5.81	5.52	10	5.66	count	8,213

The line rises at every step from the least deprived tenth to the most deprived, and the last two steps are the largest. Each point is the mean number of high conditions in a tenth of tracts on the ADI, weighted by adults, with a 95% interval from a bootstrap over counties. The second view plots the share of adults living in tracts that are high on 5 or more of the nine. Values in Table 5.

Note: PLACES values are modeled estimates. A model fitted to survey responses is applied to each tract's census population, so every condition's estimate depends on the same age, sex, race, ethnicity and education inputs and on county and state effects, and correlations across tracts are partly built in. The tract file carries no intervals, and the intervals shown here come from a bootstrap over counties. The ADI is built from American Community Survey data that overlap the model's inputs, so part of each association with it is mechanical.Note: Tenths hold equal numbers of tracts except where tied ranks fall in one tenth. Values are crude prevalence among adults, not adjusted for age, so the age mix of a tract moves them.Note: Across all the tracts in the comparison, weighted by adults, a tract is high on 2.0 of the nine conditions on average.Source: CDC PLACES 2024 release (BRFSS 2022) and ADI 2023; modeled estimates, population-weighted (adults in each tract)

Modeled estimates for tracts, not records of people

The prevalences are modeled estimates from CDC PLACES, whose 2024 release uses survey responses from 2022.[1] PLACES fits a model to survey responses and applies it to the census population of each tract. It gives each tract one prevalence for each condition among adults and holds no record of any person. A tract is high on a condition when its prevalence is in the highest 25% of tracts, with each tract counted once. A tract's count is the number of the nine conditions on which it is high. High blood pressure and high cholesterol are left out because the 2024 release has no tract estimate of either for Florida. Means and correlations are population-weighted, so each tract counts in proportion to its adult residents. In surveys that ask each adult about each condition, multiple chronic conditions means one person having more than one of them,[2] and a tract estimate cannot be turned into that count. A tract can be high on none of the nine while some of its residents have several. The report gives every estimate with its interval and sample size.

The overlap is smaller inside the most deprived tenth

Across all tracts, the average correlation between the prevalences of two of the nine conditions is +0.49 (95% CI 0.48 to 0.51). A positive correlation means that tracts high on one condition tend to be high on the other. The conditions do not go together more closely where deprivation is higher. Inside the least deprived tenth of tracts the mean correlation among the nine is +0.40, and inside the most deprived tenth it is +0.34. The count of high conditions rises with deprivation, while the overlap among the conditions inside a tenth is smaller in the most deprived tenth than in the least deprived.

Approximating the census inputs leaves most of the overlap

Part of every correlation among the PLACES estimates is built into the estimates. The model uses the age, sex, race, ethnicity and education of each survey respondent, a county measure of poverty, and state and county effects. Conditions that are modeled from the same inputs move together wherever the inputs move, whatever the residents do.

The data cannot take the model apart, but the ACS can approximate its inputs. For each condition, a weighted regression on census shares of tract residents by age, race, ethnicity and schooling, with a county poverty rate and state effects, approximates those inputs. It reproduces 79% to 90% of the variance of tract prevalence, counting the part the state effects carry. The fit leaves out the tracts that lack a census share, among them the tracts of the District of Columbia. On the tracts it keeps, the average correlation is +0.49 (95% CI 0.48 to 0.51) for the conditions themselves and +0.47 (95% CI 0.46 to 0.48) for what the fit leaves unexplained (Figure 2).

Figure 2

An approximation of the model's census inputs leaves the average correlation among the nine conditions at +0.47, against +0.49 in the raw data

Mean of the 36 pairwise population-weighted correlations among the nine conditions across 82,781 tracts, 2022, under seven comparisons

Compared within groups of tracts
After approximating the model's inputs

Show data

Data for An approximation of the model's census inputs leaves the average correlation among the nine conditions at +0.47, against +0.49 in the raw data
conf_high	conf_low	estimate	group	n	stage
0.511	0.477	0.495	Compared within groups of tracts	82,781	All tracts
0.498	0.471	0.484	Compared within groups of tracts	82,781	Within states
0.435	0.418	0.427	Compared within groups of tracts	82,781	Within counties
0.377	0.346	0.361	Compared within groups of tracts	82,781	Within tenths of the ADI
0.476	0.460	0.468	After approximating the model's inputs	82,578	Census shares, county poverty and state effects removed
0.521	0.511	0.516	After approximating the model's inputs	82,578	Census shares removed within counties
0.469	0.441	0.455	After approximating the model's inputs	82,578	Flexible fit (trees) on the same inputs and state

All seven dots stay right of zero, and the three rows that remove an approximation of the model's inputs lie close to the raw value. Each dot is the mean of the pairwise correlations among the nine conditions after the stated step, with a 95% interval from a bootstrap over counties that does not refit the adjustment. The row for all tracts and the three adjusted rows are drawn in colour, and the other rows compare tracts within states, counties and tenths of the ADI. The fits are an approximation of the PLACES model and not the model. Values in Table 6.

Note: The approximation regresses each condition on 8 ACS tract shares (ages 65 and over and under 18, white, Black, Asian and Hispanic residents, and adults with a college degree and with no more than high school), with squares and pairwise products, and removes state or county effects. It uses a county poverty rate in place of the model's measure of county poverty. The ACS file has no sex shares, and the model also uses fitted county and state effects for each condition. The fit explains 79% to 90% of the variance of each condition, counting the state effects, and the tree fit 75% to 90% out of sample. With county effects the fit explains 62% to 85% of the variance within counties. The three adjusted rows leave out the 203 tracts that lack an ACS input, among them the 201 tracts of the District of Columbia, whose ACS file lacks three of the shares; on the tracts they use, the raw average correlation is +0.49.Note: PLACES values are modeled estimates. A model fitted to survey responses is applied to each tract's census population, so every condition's estimate depends on the same age, sex, race, ethnicity and education inputs and on county and state effects, and correlations across tracts are partly built in. The tract file carries no intervals, and the intervals shown here come from a bootstrap over counties.Source: CDC PLACES 2024 release (BRFSS 2022), ADI 2023 and ACS 2019-2023; modeled estimates, population-weighted (adults in each tract)

These numbers show that the inputs which can be measured here do not remove the overlap. They do not show that the overlap is real. The fits approximate the model and are not the model. The ACS file has no sex shares, and the model works from individual respondents and not tract shares. Whatever the conditions share and the fit misses stays in what is left unexplained and keeps it correlated. The data cannot say how much of the shared part comes from the model's other inputs and how much from the neighborhoods.

A check from death records

Death records give a check that does not depend on the PLACES model. USALEEP life expectancy at birth, built from deaths in 2010 to 2015, is 6.1 (95% CI 5.8 to 6.3) years shorter in tracts that are high on eight of the nine conditions than in tracts that are high on none. Life expectancy at birth is built from death rates at each age, so the age mix of a tract does not move it the way it moves crude prevalence. That makes it a check on whether the count reflects more than age. Tracts that are high on all nine have a longer life expectancy than tracts that are high on eight, and the data here do not say why.

Figure 3

Life expectancy at birth is 6.1 years shorter in tracts that are high on eight conditions than in tracts that are high on none

Population-weighted mean tract life expectancy at birth by number of high conditions, 77,934 tracts, deaths in 2010 to 2015, institutional tracts dropped

Show data

Data for Life expectancy at birth is 6.1 years shorter in tracts that are high on eight conditions than in tracts that are high on none, Life expectancy (years)
conf_high	conf_low	count	estimate	measure	n_units
80.37	79.98	0	80.18	level	26,431
80	79.64	1	79.82	level	15,948
78.50	78.18	2	78.33	level	8,632
78.19	77.78	3	77.97	level	5,599
77.32	76.79	4	77.06	level	4,541
76.72	76.18	5	76.45	level	4,162
76.38	75.96	6	76.16	level	4,602
74.84	74.35	7	74.59	level	4,248
74.30	73.94	8	74.12	level	2,898
75.41	74.93	9	75.17	level	873

The line falls from the tracts where no condition is high to the tracts where eight are high, and ends higher at nine. Each point is the mean tract life expectancy at birth for tracts with that number of high conditions, weighted by population, with a 95% interval. Life expectancy comes from death records for 2010 to 2015 and the count from the 2022 PLACES estimates. The second view subtracts each tract's county mean to compare tracts within counties. Values in Table 8.

Note: Tract life expectancy covers 2010 to 2015 and the conditions cover 2022, so the two describe different years. USALEEP estimates were area-weighted from 2010 tracts onto 2020 tracts and suit comparisons of groups of tracts, not single tracts. The within-county view leaves out counties with a single USALEEP tract in the sample. Connecticut tracts carry planning-region codes in PLACES, which serve as its counties for the bootstrap.Note: Within a county the count varies only through the composition of tract residents, which also moves life expectancy. Life expectancy is not affected by the age mix of a tract the way crude prevalence is, which makes it a check on whether the count reflects more than age. The count itself comes from PLACES modeled estimates and carries the caveats in the notes to the earlier figures.Source: CDC PLACES 2024 release (BRFSS 2022), USALEEP life expectancy 2010-2015 (area-weighted onto 2020 tracts) and ACS 2019-2023; modeled estimates, population-weighted (ACS tract population)

Within a county the count varies only through the composition of tract residents, which also moves life expectancy, so the comparison within counties does not isolate the conditions. The two sources cover different years, 2010 to 2015 and 2022, and a correlation between them does not show that the conditions shorten lives. Pooled over tracts, the count of high conditions in the 2022 PLACES estimates rises with deprivation and goes with shorter life expectancy in the death records of 2010 to 2015. These are descriptions of places, and a count among tracts is not any resident's risk.

What this cannot tell you

No figure here says how many conditions an adult has, and reading the tract overlap as people with several conditions at once is ecological inference. The tract file has no intervals, so the bands describe the sampling of counties and not the model's own uncertainty. Values are crude prevalence among adults, so the age of a tract's residents is part of every pattern. A tract that is high on a condition is high compared with other tracts and not by any clinical standard, and the cutoff moves every count.

Go deeper

The report: every estimate with its interval and sample size.
Tables: every figure of the report as a table.

Data sources

CDC PLACESCDC PLACES 2024 release, census tracts, crude prevalence among adults of nine conditions (BRFSS 2022)Years: 2024 release (tracts); 2025 release (ZCTAs, 2023 data year) · Weights: adult population of the tract · License: Public domain (CDC, U.S. Department of Health and Human Services)
Area Deprivation IndexNeighborhood Atlas Area Deprivation Index, 2023, national percentileYears: 2023 release (version 4.0.1; built on ACS 2019-2023) · Weights: adult population of the tract · License: Free for non-commercial use; cite the Neighborhood Atlas and Kind and Buckingham
American Community SurveyACS 2019-2023 tract shares used to approximate the model's inputsYears: 2019-2023 (5-year) · Weights: adult population of the tract · License: Public domain (U.S. Census Bureau)
USALEEP life expectancyNCHS small-area life expectancy at birth, 2010-2015, 2020 tractsYears: 2010-2015 · Weights: ACS 2019-2023 tract population · License: Public domain (NCHS, CDC)
Published estimates from cited studiesEstimates quoted in the causal box (Ludwig and coauthors 2011)Years: as reported in each study · Weights: as reported in the study · License: Quoted from published research for commentary

All data sources

References

Greenlund, Kurt J.; Lu, Hua; Wang, Yan; Matthews, Kevin A.; LeClercq, Jennifer M.; Lee, Benjamin; Carlson, Susan A. (2022). PLACES: Local Data for Better Health. Preventing Chronic Disease 19: 210459. doi:10.5888/pcd19.210459
Ward, Brian W.; Schiller, Jeannine S. (2013). Prevalence of Multiple Chronic Conditions Among US Adults: Estimates From the National Health Interview Survey, 2010. Preventing Chronic Disease 10: 120203. doi:10.5888/pcd10.120203

How to cite

@misc{singh2026multiplechronicconditions,
  author = {Singh, Vishal},
  title = {Multiple chronic conditions},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/multiple-chronic-conditions},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}