Moving to greener neighborhoods and healthcare costs in Northern California

Abstract

Background:

Environmental drivers of healthcare costs are mainly unknown. Studies that account for changes in longitudinal, within-individual changes accompanying changes in residential addresses have not been conducted. In the current study, we tested whether moving to a greener neighborhood was associated with decreases in healthcare costs and whether these associations vary by individual and neighborhood-level characteristics.

Methods:

Cohort study with 13-year administrative data from Kaiser Permanente Northern California members who moved (N = 129,576) and did not move (N = 4,807,135). Healthcare costs and neighborhood characteristics before and after residential moves were examined, considering age, sex, race, neighborhood socioeconomic status, air pollution, and population density. Change in neighborhood greenery was measured as changes in deciles of satellite-derived Normalized Difference Vegetation Index (NDVI) in 250-m, 500-m, and 1000-m Euclidean buffers. Nonmovers were retained for analyses as they represented individuals with no changes in neighborhood greenery.

Results:

We found that moving to greener neighborhoods was generally not associated with changes in healthcare costs. However, moving to greener areas was associated with decreases in outpatient costs (NDVI-250-m relative cost ratio [RCR] = 0.993, 95% confidence interval [CI] = 0.987, 0.999). Movers with the greatest increase in greenness had $89 less annual outpatient costs than movers who maintained their greenness levels. Subgroup analyses found significant decreases in outpatient costs for women (RCR = 0.992, CI = 0.985,0.999), adults aged 18–44 (RCR = 0.989, CI = 0.981,0.996), and individuals moving from lower population density-neighborhoods (RCR = 0.991, CI = 0.983,0.999).

Conclusions:

This longitudinal, within-individual analysis found little evidence to support the hypothesis that moving to greener neighborhoods is associated with reduced healthcare spending. Future studies should incorporate data on duration of residence, individual socioeconomic status, reasons for moving, and broader economic outcomes to better evaluate how residential environments may influence healthcare costs.

What this study adds

We report the first within-individual study of whether changes in residential greenness affect healthcare costs. We found little evidence to support the hypothesis that moving to greener neighborhoods reduces healthcare costs, apart from a modest reduction in outpatient costs. These findings were derived from a large sample size of nearly 130,000 movers across 13 years from diverse ecological and urban-rural gradients, suggesting our results are robust and associations between residential greenness and healthcare costs are complex and inconsistent.

Introduction

The health benefits of greenspace are increasingly recognized. Observational and experimental studies suggest that greater residential greenspace is associated with reduced risks of chronic disease, improved mental health, and lower all-cause mortality.¹ Correspondingly, researchers have begun to examine whether these health benefits translate into lower healthcare costs.²

Only a handful of studies have investigated the relationship between greenspace and healthcare spending using individual-level data. Prior work using the Kaiser Permanente Northern California (KPNC) administrative data found that greater residential greenness was associated with lower healthcare costs, with particularly strong associations observed in high-cost categories such as inpatient and pharmacy costs.³ Other studies in Australia have shown that more tree canopy is associated with reduced hospitalization costs for cardiovascular events.⁴ These studies, however, are largely cross-sectional or rely on between-person comparisons, which are vulnerable to confounding by stable individual differences and contextual factors.

A within-individual, longitudinal design—such as one tracking people who move between neighborhoods—offers some advantages. By comparing each individual to themselves over time, such designs inherently control for time-invariant personal characteristics (e.g., baseline health status, genetic risk, and long-standing behaviors), even if those traits are unmeasured. They also reduce bias due to geographic self-selection and unobserved between-person differences.⁵ Overall, this approach enables a more rigorous test of whether changes in residential greenspace are associated with changes in healthcare costs.

To our knowledge, no prior study has applied this within-individual design to examine whether residential moves to greener neighborhoods lead to reductions in healthcare spending. To address this gap, this study aimed to conduct the first longitudinal, within-individual analysis of changes in greenspace and healthcare costs, using 13 years of administrative data from KPNC.

Methods

Study population

We conducted this study using membership data from KPNC, a large, comprehensive, integrated healthcare system serving a diverse population. The study population included 4,936,711 members with residential addresses recorded in the administrative database, geographically located within our 22-county study area (approximately 200,000 km²), and residing at that address for at least 2 years before or within our 13-year study period (2003–2015). Of these, 129,576 individuals moved at least once during the study period and had complete data for at least two consecutive years. Individuals with multiple addresses were classified as movers, while those with the same address throughout were classified as nonmovers (N = 4,807,135).

We retained both movers and nonmovers in the analytic sample. Nonmovers and movers with no change in Normalized Difference Vegetation Index (NDVI) decile served as the referent group for comparisons with individuals who experienced increases or decreases in greenspace exposure. Including all members in the sample allowed us to use a within-individual change design while leveraging the full data to model healthcare costs across exposure gradients.

On enrollment in the health plan, all KPNC members were informed that their data may be used for research. Informed consent was waived by the Kaiser Permanente Foundation Research Institute institutional review board.

Covariates

The self-reported age, sex, and race/ethnicity of members were drawn from electronic health records and administrative databases. The Neighborhood Deprivation Index,⁶ which captures geographic variation in poverty, education, employment, housing stock, and occupation for each census block group, was calculated based on eight U.S. Census variables. Population density was also assessed at the census block group level. Air quality data, specifically census tract level PM2.5 for 3 years (2012–2014), were drawn from the CalEnviroScreen 2.0.⁷ Changes in neighborhood characteristics were calculated for each relocation (postmove minus premove) and were time-varying in individuals with multiple relocations. Age was also time-varying, while other demographic characteristics (sex, race/ethnicity) were static.

Healthcare costs

Healthcare costs for each member for 2003–2015 were compiled from KNPC administrative and billing records. These comprised services received both within and outside of the KPNC system, provided they were reimbursed through the KPNC network. All costs were adjusted to 2016 dollars using the Consumer Price Index. Because gamma distribution modeling would exclude records with no costs, we added $1 to each category of summarized costs each year for each individual in the study. Per-patient direct healthcare costs in each calendar year of follow-up were calculated, accounting for variable time in the cohort for each month of each study year to accommodate individuals having differing cohort entry and exit points.

Residential greenspace

We estimated residence greenspace exposure with the NDVI in Euclidean (straight-line) buffers around each member’s residential street address(es). NDVI indicates the amount of healthy green vegetation (“greenness”) in each pixel based on the reflection of near and far infrared bands of light. NDVI values range from −1.0 (water/ice/snow) to 0.0 (built environment) to 1.0 (complete healthy vegetative cover). The primary analyses examined 250-m buffer based on our previous findings for residential greenspace and healthcare costs in earlier work. Sensitivity analyses also examined larger buffer sizes of 500 and 1000 m.³

We used Google Earth Engine to assemble mean NDVI values from satellite images collected every 8 days in all 12 calendar months from 2003 to 2015. The code used to assemble the images and calculate NDVI values was adapted from an International Society of Environmental Epidemiology training⁸ and is provided in Appendix A and Appendix B in the Supplementary Materials. Two satellites were included based on image availability: Landsat 5 for 2003–2013 and Landsat 8 for 2014–2015. The spatial resolution was 30 m × 30 m. In line with past research, we reclassified negative NDVI values as missing data to prevent water bodies from downgrading greenspace values since water has also been associated with some health benefits.⁹

NDVI values were categorized in deciles to allow the greenness-cost relationship to be nonlinear over levels of greenness. The ranges of each decile were determined by examining NDVI values across all years and all individuals in the cohort, agnostic of changes in residential address. Changes in NDVI were calculated by subtracting the decile postmove from the decile premove. Deciles were chosen to account for possible nonlinear relationships between the exposure and outcome. Decile ranges were calculated by examining NDVI values for the entire sample for all study years. Changes in NDVI deciles were calculated by subtracting the decile level after the move from the level before the move.

Analyses

To account for within-person correlation in repeated annual measures of healthcare costs over the study period, we used generalized estimating equations with an autoregressive covariance structure. This approach allowed us to model changes in healthcare costs while accommodating differing lengths of follow-up and multiple observations per individual. Costs were calculated for each calendar year in which a member had KPNC coverage. For movers, this included years before and after each residential move.

We estimated models at two levels of covariate adjustment. Model 1 was minimally adjusted and included calendar year, intercept, age, sex, race/ethnicity, baseline greenness decile, and change in greenness decile. Model 2 included additional adjustments for neighborhood-level characteristics: baseline and change scores in deprivation (Neighborhood Deprivation Index), population density, and PM2.5.

Models were run for total healthcare costs as well as four cost subcategories: outpatient, inpatient, emergency room, and pharmacy (i.e., prescription medications dispensed at KPNC outpatient pharmacies and their acquisition costs). In addition to main effects models, we conducted exploratory stratified analyses by sex, age group, neighborhood deprivation, and urbanicity, based on evidence that these factors may modify associations between greenspace and health.^10–13

To test for potential spatial clustering that could bias estimates, we calculated Moran’s I for the midpoint year of the study within a 16-km buffer around each member’s residence. No spatial autocorrelation was detected (Moran’s I < 0.0001).¹⁴ All analyses were conducted using SAS Version 9.4 (Cary, NC).

Results

Approximately 3% (N = 129,576) of the study population moved at least once during the 13-year study period (Table 1). Among movers, only 1% (N = 1,322) moved twice or more times. Movers were predominantly aged 18–44 years and tended to reside in neighborhoods with higher levels of deprivation and population density, relative to nonmovers. Median total direct costs were lowest among <18-year-olds; twice as high among 18–44-year-olds as <18-year-olds; and three times as high among ≥45-year-olds as <18-year-olds (Table S1; https://links.lww.com/EE/A347). Total mean costs tended to increase across the study period but not linearly (Figure S1; https://links.lww.com/EE/A347). Inpatient costs had the highest annual mean value, and outpatient and emergency room costs had lower annual mean values (Table S2; https://links.lww.com/EE/A347).

Table 1.

Characteristics of the sample

	Total				Nonmovers				Movers
	N = 4,936,711				N = 4,807,135				N = 129,576
Individual level	N (%)				N (%)				N (%)
Sex
Male	2,522,817 (51.10)				2,456,410 (51.10)				66,407 (51.25)
Female	2,413,894 (48.90)				2,350,725 (48.90)				63,169 (48.75)
Age
<18 yrs	1,423,397 (28.83)				1,384,676 (28.80)				38,721 (29.88)
18–44 yrs	1,940,424 (39.31)				1,865,930 (38.82)				74,494 (57.49)
≥45 yrs	1,572,890 (31.86)				1,556,529 (32.38)				16,361 (12.63)
Race/ethnicity
White	2,109,993 (42.74)				2,058,232 (42.82)				51,761 (39.95)
Black	340,120 (6.89)				326,883 (6.80)				13,237 (10.22)
Asian	748,927 (15.17)				729,843 (15.18)				19,084 (14.73)
Hispanic	935,748 (18.95)				903,168 (18.79)				32,580 (25.14)
Other/unknown	801,923 (16.24)				789,009 (16.41)				12,914 (9.97)
Neighborhood level	Med	IQR	Min	Max	Med	IQR	Min	Max	Med	IQR	Min	Max
Greenness (NDVI-250m)
Premove	0.22	0.08	0.02	0.61	0.22	0.08	0.02	0.61	0.22	0.07	0.02	0.58
Postmove	0.22	0.08	0.02	0.61	-	-	-	-	0.22	0.07	0.02	0.54
Change	0.00	0.00	−0.47	0.45	-	-	-	-	0.00	0.09	−0.47	0.45
Deprivationa
Premove	−0.25	1.09	−2.12	4.23	−0.25	1.09	−2.12	4.23	−0.09	1.18	−2.12	4.23
Postmove	−0.25	1.08	−2.12	4.23	-	-	-	-	−0.20	1.04	−2.12	4.23
Change	0.00	0.00	−4.57	4.64	-	-	-	-	0.00	0.90	−4.57	4.64
Population density
Premove	6134	6339	0	161,499	6127	6331	0	161,499	6571	6099	2	16,1499
Postmove	6127	6302	0	161,499	-	-	-	-	5987	5867	2	142,920
Change	0	0	–159,900	142,899	-	-	-	-	0	5790	−159,900	142,899
Air pollution
Premove	10.56	0.96	6.81	15.87	10.56	0.96	6.81	15.87	10.67	0.89	6.81	15.87
Postmove	10.56	0.96	6.81	15.87	-	-	-	-	10.69	0.88	6.81	15.87
Change	0.00	0.00	−8.72	8.12	-	-	-	-	0.00	0.45	−8.72	8.12

Change scores are calculated as post minus pre.

IQR indicates interquartile range; Med median.

^aHigher values indicate more neighborhood deprivation (lower socioeconomic status).

The distribution of changes in greenness was relatively normal with a peak at the midpoint (NDVI = 0.00) (Figure S2; https://links.lww.com/EE/A347). NDVI values tended to change by only ±0.02 for each unit change in greenness decile (Tables S3–S5; https://links.lww.com/EE/A347). Exceptions were observed at the extremes of the range. An 8-decile increase represented a net gain in NDVI of approximately 0.06, whereas a 9-decile increase represented a net gain in NDVI of approximately 0.23. This meant that a mover who experienced the greatest decile increase in our sample witnessed a median increase in NDVI_250m of 0.24 (0.17–0.43).

Minimally adjusted models found no associations between greenness and total, inpatient, or emergency room costs (Table 2 and Tables S6–S8; https://links.lww.com/EE/A347). By contrast, significant reductions in outpatient costs in unadjusted models were observed (Model 1), but adjusting for deprivation, air pollution, and population density attenuated the associations (Model 2). In this case, only increases in greenness within the smallest area around the home (250-m buffer) were associated with reduced outpatient healthcare costs. Specifically, movers who experienced the largest increase in residential greenness—shifting from the lowest to highest NDVI decile—had an estimated $89 lower annual outpatient cost compared to those with no change in greenness. For reference, this nine-decile increase corresponded to a median NDVI-250 increase of 0.24 (Table S3; https://links.lww.com/EE/A347). While pharmacy data existed in the data, models examining only this category of costs did not converge and were excluded.

Table 2.

Association between changes in residential greenness and healthcare costs

	Model 1		Model 2
	RCR	95% CI	RCR	95% CI
Total costs
Change in NDVI-250	0.995	0.989, 1.001	0.997	0.989, 1.004
Change in NDVI-500	0.995	0.989, 1.002	0.998	0.990, 1.005
Change in NDVI-1000	0.997	0.991, 1.004	1.001	0.993, 1.009
Inpatient costs
Change in NDVI-250	0.999	0.984, 1.013	1.012	0.992, 1.033
Change in NDVI-500	1.001	0.986, 1.016	1.003	0.984, 1.022
Change in NDVI-1000	1.008	0.993, 1.023	0.999	0.982, 1.017
Outpatient costs
Change in NDVI-250	0.992	0.986, 0.999	0.993	0.987, 0.999
Change in NDVI-500	0.992	0.986, 0.999	0.994	0.987, 1.000
Change in NDVI-1000	0.992	0.985, 0.999	0.993	0.986, 1.000
Emergency room costs
Change in NDVI-250	1.000	0.995, 1.006	0.999	0.993, 1.005
Change in NDVI-500	0.999	0.993, 1.004	0.999	0.993, 1.006
Change in NDVI-1000	0.997	0.991, 1.002	1.000	0.993, 1.007

Change scores are calculated as postmove minus premove decile of greenness. Model 1 adjusted for baseline decile of greenness, year, intercept, and individual-level characteristics, including age, sex, and race/ethnicity. Model 2 adjustments add neighborhood-level characteristics, including baseline levels and change scores for deprivation, air pollution, and population density.

CI indicates confidence interval; RCR, relative cost ratio (relative mean costs per year among those in one unit change group for NDVI relative to mean costs of those with no changes in NDVI).

Figure 1 summarizes the associations between changes in greenness and healthcare costs in subgroups of movers. We found that greenness increases were associated with lower outpatient costs among women, individuals 18–44 years old, and individuals from less dense neighborhoods. Individuals from less dense neighborhoods also showed lower total healthcare costs among movers with increases in greenness. No other subgroups showed significant associations with any type of cost (Tables S9–S12; https://links.lww.com/EE/A347).

Figure 1.

Subgroup-specific relative healthcare cost ratios by changes in deciles of residential greenness (NDVI-250 m). ER indicates emergency room, models adjusted for baseline greenness decile, year, intercept, sex, age, race/ethnicity, neighborhood deprivation, air pollution, and population density.

Discussion

The current study provides the first longitudinal, within-individual-change examination of whether moving to a greener neighborhood influences healthcare costs. We found little evidence to support this possibility. Changes in residential greenness did not influence total, inpatient, or emergency room costs. However, we observed that moving and increasing residential greenness was associated with a modest reduction in outpatient costs, including routine or specialist doctor visits and same-day procedures (i.e., colonoscopy, endoscopy, and minor surgeries). These protective effects persisted in fully adjusted models when measuring greenness very closely around the home (250-m or a 3-minute walk). When stratifying the sample, the protective effects of changes in residential greenness on outpatient costs were only observed in residents who were female, 18–44 years old, or coming from lower population densities.

Differences between our findings and past research in California³ and Australia⁴ could be attributable to contrasting lengths of residence and age. Movers to greener areas were considerably younger than their nonmover counterparts. Residential greenspace estimates may represent longer term, cumulative exposures, while data from movers may represent shorter term exposures that revealed effects on less serious outpatient services. Another interpretation is the dominance of people aged 18–44 in our sample of movers. This age bracket had lower healthcare costs and lower residential greenness values than older adults in our sample (Table S1, Figure S3; https://links.lww.com/EE/A347). Prior research shows healthcare costs begin to increase sharply after age 45, rising approximately 2.7-fold for females and 4.4-fold for males by ages 85–89, primarily due to the increased incidence of chronic conditions and age-related illnesses.¹⁵ The potential for outpatient costs to be buffered by greenspace may have been obscured by the relatively low share of middle-aged and older adults in our sample with longer residential histories and higher greenspace levels. Additionally, in contrast to another greenspace study of prescription and talking therapy referral costs,¹⁶ we examined costs for all conditions and illnesses, including physical health. Restricting economic estimates to mental health benefits of greenspace overlooks other costly conditions, including diabetes, heart disease and hypertension, chronic pain, injury by falling, and cancer.^17–19

The potential association between greenspace and healthcare costs could be explained through multiple mechanisms.^20–22 Greenspace exposure can improve health status, which may influence healthcare utilization and then healthcare costs.^2,3 For instance, air pollution impacts physical health due to oxidative stress, genotoxicity, harmful inflammation, cellular death, and changes to the gut microbiome.²³ Some types of vegetation may reduce air pollution by providing surfaces—such as foliage—on which gases and particulates can be deposited; however, the overall contribution of this mechanism to improved air quality remains debated, with some evidence suggesting the effects are modest or context-dependent.²⁴ Similarly, vegetation can mitigate other environmental stressors associated with poor health, including traffic noise, urban heat islands, and artificial light at night.²⁵ Greenspace can also promote recovery from chronic stress and attention fatigue.^26,27 Time outdoors in nature can also promote social interaction and reduce loneliness,²⁸ physical activity,²⁹ and sleep,^30,31 which are strong predictors of cardiovascular health.³² As shown previously, comorbidities, body weight, and smoking behavior attenuate the effect estimates between greenspace and healthcare costs, supporting a hypothesized causal pathway that greenspace impacts costs through direct effects on health status.³

Strengths and limitations

A strength of our study lies in our large, diverse population within a single healthcare system, which provides more uniform access to care and helps reduce potential disparities in healthcare utilization. While we did not stratify analyses by all dimensions of diversity, its presence enhances the generalizability of findings to a broader U.S. population. By using a within-individual design that evaluates changes in greenspace exposure following residential moves, we were also able to adjust for time-invariant individual characteristics.

However, we acknowledge several limitations. The absence of a direct measure of individual health status may result in residual confounding, and self-selection into neighborhoods with more or less greenspace remains a possibility.⁵ Additionally, while healthcare costs provide a broad and policy-relevant endpoint, they are influenced by numerous factors and should not be interpreted as a direct proxy for health outcomes. Neither could we account for individual socioeconomic status, which can influence the quality of residential greenspace and confound the effects of greenspace quantity on health.³³ Greenness was also measured with a measure of quantity and lacked detail on types of vegetation, quality of greenery, or whether people spend time outdoors in green spaces. Another limitation was the lack of information on how long participants remained at their new address following a move. Without a minimum residence duration requirement, it is possible that some individuals did not experience sufficient exposure to the new neighborhood environment for meaningful changes in healthcare costs to occur. Given the number of cost categories, buffer sizes, and subgroup analyses examined, the potential for false positives due to multiple comparisons also cannot be ruled out.

Further research examining movers and changes in greenspace and healthcare costs is warranted—particularly studies that incorporate data on individual-level socioeconomic status, reasons for moving, and other relevant contextual factors. Additional work should also explore broader economic outcomes, including costs beyond formal healthcare, such as informal caregiving and lost productivity.

Conclusion

This investigation of residential moves and resulting changes in greenspace and healthcare costs found largely null results. This contrasts with earlier work in the same population, which identified associations between higher levels of residential greenness and lower healthcare costs. In this movers-only analysis, increasing greenness was tied to modest reductions in outpatient costs, but no other statistically significant associations were observed. Taken together, these findings suggest that greenspace might influence healthcare costs through cumulative or longer term exposures.

Conflicts of interest statement

The authors declare that they have no conflicts of interest with regard to the content of this report.