Novel hot and cold spots of young-onset colorectal cancer mortality in United States counties

Young-onset colorectal cancer (yoCRC) incidence and mortality have been on the rise in the U.S. for decades¹ but have only recently begun to gain public health attention.² Causes are largely unknown, but modifiable factors are of particular concern due to a strong birth cohort effect, where each subsequent cohort of births since the 1950s has higher CRC risk.³ Putative risk factors and prognostic indicators for yoCRC include smoking, obesity, alcohol consumption, diabetes, sex, race, and socioeconomic status.

Geospatial patterns of yoCRC mortality rates have received little attention but are important for understanding the underlying drivers of mortality and allocating public health resources. Prior geographic clustering studies were limited by their descriptive designs, exclusive focus on hot spots, and lack of cluster-specific relative risks (RRs).^4,5 Adjustment for known risk factors is critical for determining which hot and cold spots represent areas where anomalous mortality rates exist. This study was designed to robustly identify hot and cold spots of yoCRC mortality using county-level data, with the goal of highlighting regions for further investigation and future preventive interventions.

Using discrete Poisson circular spatial scans in SaTScan™, version 10.0.1, we identified adjusted hot and cold spots of aggregated 1999–2019 CRC mortality in 3,036 U.S. counties for two broad age-at-diagnosis categories, <50 and 50+, along with four stratified categories: <35, 35–49, 50–64, and 65+ (Supplementary Methods).⁶ Mortality rates from CDC WONDER Underlying Cause of Death data were stabilized via spatial empirical Bayes smoothing, and adjusted quasi-Poisson (QP) spline models were fit. The adjusted models included county-level median age, sex, race/ethnicity, area deprivation index (ADI), obesity, smoking, and binge/heavy drinking. Resulting RRs signified clusters where observed yoCRC deaths were more or less common than expected yoCRC deaths, proportional to cluster population size.

In 1999–2019, there were 69,976 U.S. CRC deaths among those diagnosed before age 50 (7,326 <35). This equated to a mean smoothed county yoCRC mortality rate of 1.78 deaths per 100,000 population (SD=0.54; range=0.19–5.34 deaths). For the same period, there were 1,033,541 CRC deaths among those diagnosed at 50+ years, equivalent to a mean smoothed average-onset CRC (aoCRC) mortality rate of 56.82 deaths per 100,000 population (SD=10.34; range=12.19–102.36). Counties had these averages: a median age of 40.1, 50.1% female, 9.1% Black, 8.4% Hispanic, 1.2% Asian, ADI of 100.1, 19.4% current smokers, 16.8% binge/heavy drinkers, and 33.7% obese.

All predictors in adjusted QP spline models were significantly associated with yoCRC mortality; overall and within finer age strata. Directions of association remained consistent for all predictors except for binge/heavy drinking, which was positively associated with yoCRC mortality in the youngest stratum (<35) but had a negative association in the 35–49 and <50 strata (results not shown). Diagnostics showed that all adjusted spline models had good fit, namely via excellent deviance R² values of 0.29, 0.45, 0.60, 0.68, 0.53, and 0.70 for the <35, 35–49, <50, 50–64, 65+, and >50 age-at-diagnosis strata, respectively.

Adjusted scan results for the <35 group revealed two significant mortality hot spots constituting the northeast (RR: 1.25, p<0.0001) and upper Midwest (RR: 1.11, p<0.0001), and three significant cold spots constituting the southwest (RR: 0.74, p<0.0001), California (RR: 0.78, p<0.0001), and mountain west (RR: 0.82, p<0.0001) (Figure 1, panel D). Adjusted scan results for ages 35–49 identified three significant hot spots constituting the southeast (RR: 1.20, p<0.0001; RR: 1.16, p<0.0001) and Great Lakes region (RR: 1.12, p<0.0001) and five significant cold spots constituting the pacific/mountain west (RR: 0.90, p<0.0001), California (RR: 0.82, p<0.0001), southern Texas (RR: 0.89, p<0.0001), and the southwest more broadly (RR: 0.86, p<0.0001) (Figure 1, panel F). Adjusted scan results for yoCRC overall (age <50) revealed two hot spots constituting the southeast (RR: 1.24, p<0.0001) and Great Lakes region (RR: 1.10, p<0.0001) and four cold spots constituting lower Wisconsin (RR: 0.87, p=0.0002), the northeast (RR: 0.92, p=0.012), southwest Texas (RR: 0.90, p<0.0001), and western counties more broadly, including Alaska (RR: 0.82, p<0.0001) (Figure 1, panel B). Figure 1 panels A, C, and E show unadjusted yoCRC scan results. Though cold spots were similar across strata, yoCRC hot spots shifted southward in the 35–49 age stratum in comparison to the <35 group. aoCRC scan results display that southeastern mortality clusters are also a pattern among older age strata (Supplementary Figure 1).

Figure 1. Unadjusted and adjusted young-onset colorectal cancer Poisson spatial scan clusters and mortality rates per 100,000 U.S. population.

(A, C, E) Unadjusted. (B, D, F) Adjusted for county median age, percent female, race/ethnicity, area deprivation, current smoking, binge or heavy alcohol consumption, and obesity. The spatial scan clustering circles include statistically significant Gini clusters (p<0.05) with cluster populations ≥2.5% and ≤25% of the age-specific total population, where clusters represent areas in which observed deaths were significantly more or less common than expected deaths, proportional to cluster population size.

Our study revealed new mortality hot spots in the Midwest/northeastern Great Lakes region, which constitutes the first reporting of yoCRC mortality hot spots in these regions. Further, after covariate adjustment, mortality hot spots remained in southern and Appalachian counties in the <50 age group, reflecting previous unadjusted analyses.^4,5 However, after stratifying yoCRC mortality into younger (<35) and older (35–49) age-at-diagnosis categories, 35–49 hot spots more closely followed southern patterns seen in aoCRC, while <35 hot spots did not. Therefore, deaths among the youngest yoCRC patients may be driven by a distinct set of factors (e.g. more aggressive biology) as compared to deaths among older yoCRC and aoCRC patients. Future studies focused on risk stratification for individuals below the current CRC screening age of 45 years should incorporate geographic factors.⁷ Our study is also the first exploration of yoCRC mortality cold spots and suggests that western/southwestern counties have lower risk of yoCRC death. The presence of post-adjustment hot/cold spots indicates that unmeasured factors, such as access to care and related treatment disparities, may drive anomalous yoCRC mortality rates, either independently or in conjunction with demographic/modifiable variables accounted for here.

This study had several limitations including its ecological nature, race and time-aggregated deaths to offset data suppression, and lack of adjustment for stage-at-diagnosis due to data limitations. Histogram results for a subsample of late-stage all-age incident cases revealed that adjusting for stage-at-diagnosis would have had little effect, as 86% of subsample counties had late-stage case percentages in the 50%−70% range. Despite limitations, we discovered novel results that may initiate new research examining region-specific and age stratum-specific yoCRC mortality-related exposures (e.g diet, provider access, or antibiotic use) and provide an impetus to target interventions in particular regions. Future work should consider examining race/ethnicity-stratified and patient-level spatiotemporal clustering.

Abbreviations

ADI

area deprivation index

QP

quasi-Poisson

RR

relative risk

yoCRC

young-onset colorectal cancer

aoCRC

average-onset colorectal cancer