Multimorbidity clusters in adults 50 years or older with and without a history of cancer: National Health Interview Survey, 2018

The National Health Interview Survey (NHIS) (https://​www.​cdc.​gov/​nchs/​nhis/​) is a cross-sectional household survey of the US civilian, noninstitutionalized population conducted annually by the National Center for Health Statistics (NCHS) [24]. NHIS collected data on demographics, health-related characteristics, and multimorbidity. In 2018, the final response rate for the sample adult component was 53.1% [24]. More detail about how the sample adult component was selected can be found in the NCHS 2018 Survey Description [24].

We restricted the analysis to participants with 2 or more diseases (including cancer) and conducted an exploratory factor analysis for cancer survivors and controls. Given that this was intended to be an exploratory analysis of disease clusters and differences in clusters found in adults with cancer compared to controls focused on diseases and cancer status, and not to make nationally representative estimates of these conditions, we treated the NHIS data as a convenience sample; all analyses are unweighted and did not account for the complex survey factors. The cluster analysis used an orthogonal varimax rotation. Factors were extracted with eigenvalues > 1 and retained after rotation if the variance explained was > 5%. Items with a moderate to high loading of at least 0.3 on any factor were retained for the corresponding factor [28]. Items could potentially load on multiple factors.

Table 1 summarizes sample characteristics by cancer status. Significant differences were observed in age, sex, race and ethnicity, and smoking status between cancer survivors and controls. Controls comprised more 50–64-year-old adults (43.0%) compared to cancer survivors (25.4%). Conversely, cancer survivors had almost twice the percentage of adults aged 85 + (12.3%) compared to controls (6.4%). A greater percentage of cancer survivors were widowed/divorced/separated adults (50.2%) compared to controls (42.8%). More cancer survivors were college graduates or had some college compared to controls (60.2% vs 55.8%).

Fewer cancer survivors were current smokers (8.9%) and obese (32.0%) compared to controls (14.6% and 39.9%, respectively). There were no significant differences in distribution of self-rated health status, family income, alcohol use, or physical activity between cancer survivors and controls. The majority of cancer survivors were diagnosed more than 5 years before (63.1%). The survivor sample comprised primarily former breast cancer (47.6%) patients, and most survivors had been diagnosed at ages 50–64 (44.1%).

Table 2 summarizes disease prevalence in the sample stratified by cancer status. Controls reported a significantly higher prevalence of hypertension, heart attack, diabetes, arthritis, and high cholesterol compared to cancer survivors, whereas cancer survivors reported higher prevalence of emphysema. There was a significant difference in the distribution of the number of multimorbidities between cancer survivors and controls (Table 2). Cancer survivors had higher prevalence of reporting 4 or more diseases compared to controls (Fig. 1), whereas Table 2 reveals that controls more frequently reported multimorbidity counts of 2–3.

The cluster analysis yielded 6 clusters for cancer survivors and 4 clusters for controls. Only clusters that matched across both groups were named. These included the pulmonary cluster, with 3 conditions: COPD, emphysema, and asthma; the cardiac cluster, with 3 conditions: coronary heart disease, heart attack, and angina/other heart condition; and the liver cluster, with 2 conditions: hepatitis and liver disease. Additionally, there were 3 unmatched clusters for cancer survivors and 1 unmatched cluster for the controls.

Clusters occurring in both cancer survivors and controls were compared in Table 3. In both the pulmonary cluster and the cardiac cluster, there were significant differences between cancer survivors and controls in age group, sex, marital status, smoking status, and multimorbidity count. Significant differences in race/ethnicity were observed in the cardiac cluster only. In the liver cluster, there were significant differences in age group and multimorbidity count between cancer survivors and controls.

Table 4 displays characteristics of clusters that did not match across cancer survivor and control groups. There were 3 unmatched clusters for cancer survivors. Unmatched cluster 1 for cancer survivors contained 3 conditions: hypertension, high cholesterol, and diabetes. Unmatched cluster 2 for cancer survivors contained 2 conditions: stroke and arthritis. And unmatched cluster 3 for cancer survivors contained 2 conditions: high cholesterol and ulcer. There was 1 unmatched cluster for controls, which contained 6 conditions: hypertension, high cholesterol, arthritis, asthma, diabetes, and ulcer. Further analyses could not be conducted on these groups due to a lack of comparison group. Demographic composition of each unmatched cluster was similar to demographics for cancer survivor and control groups, as seen in Table 1.

Figure 1 demonstrates the percentage of adults with 2–3 diseases is higher for controls compared to cancer survivors. However, the percentage of adults with 4–10 diseases is higher for cancer survivors compared to controls. Table 5 demonstrates that a higher proportion of cancer survivors age 85 + had 2–3, 4–5, or 6 + diseases (10.7%, 11.8%, and 16.8%, respectively), compared to adults 85 + in the control group (5.57%, 7.56%, and 8.27%, respectively). Notably, in the age 85 + groups, there were twice as many cancer survivors with 6 + diseases (16.8%) compared to controls (8.27%).

The strengths of this study include: first, the use of NHIS data, which is drawn from the US population and includes the fee-for-service Medicare population. Second, analyses of multimorbidity clusters in older adults typically restrict samples to adults aged 65 and older. However, the majority of cancer diagnoses in our population-based sample occurred at ages 50–64, so inclusion of this age group in our analysis was imperative. Multimorbidity has a significant impact on HRQOL [38] and health care expenditures [30] in this age group. Additionally, insurance coverage is not guaranteed for adults age 50–64, which may contribute to age-related disparities in access to care. Therefore, this is a critical age group in which to focus both cancer and noncancer related preventive measures, to improve multimorbidity-related outcomes among older age groups. Third, although we were unable to include all racial/ethnic groups, the inclusion of non-Hispanic Black persons and Hispanic persons increases our understanding of differences in multimorbidity clusters, cancer diagnoses, and sociodemographic factors across race and ethnicity. Fourth, prior studies of multimorbidity clusters include cancer as a multimorbidity [22], analyze symptom clusters in relation to a specific cancer type [39], or do not include a noncancer comparison group [23]. However, significant differences between the cancer survivor and noncancer groups in our study demonstrate that cancer diagnoses are associated with higher multimorbidity overall, as well as certain multimorbidity clusters per demographic factors such as age, sex, and race/ethnicity. Finally, our analysis includes multimorbidity clusters rather than simple counts, dyads, or triads, because multimorbidity counts do not demonstrate which specific combinations of diseases are associated with health care utilization [22, 40], disability [41], or complexity [40]. Therefore, although more difficult to analyze and interpret, multimorbidity clusters can inform more focused, economical, and effective prevention strategies.

The results of a multimorbidity cluster analysis depend on how multimorbidity is defined [2] and the conditions included in the analysis [34]. Due to the lack of consistency in definitions, data sources, and methodology, Goodman et al. [42] proposed a list of 20 diseases to include in future multimorbidity studies. However, we could not utilize this proposed list because not all diseases were assessed directly or used the same question stem in the NHIS 2018 survey. Only conditions assessed using the “Have you ever been told” question stems were included in the cluster analysis, to minimize variability in multimorbidity related to recency of diagnosis. As a result, our analysis did not include the following recommended diseases: autism, chronic kidney disease, dementia, depression, HIV/AIDS, osteoporosis, schizophrenia, or substance use disorder [42]. The exclusion of mental health conditions is a significant limitation given their prevalence in the US and the association between mental health disorders, multimorbidity, and increased incidence of common diseases [43], hospital length of stay [44], and risk of care dependence [31]. Moreover, for individual diseases, NHIS does not include questions about disease severity, management, treatment methods, age at diagnosis, or resolution. Diseases in NHIS are also self-reported, and there is evidence of misalignment between self-reported multimorbidity count and multimorbidity count from other data sources, such as insurance claims, electronic health records, and reports from providers [45, 46].

Of note, we use the term multimorbidity as defined by the presence of 2 or more simultaneous diseases, rather than chronic conditions, in an individual. We also discuss individual disease or diseases, instead of the more commonly used terms—chronic conditions or multiple chronic conditions—because our and other common data sources in the literature do not include temporality of disease diagnosis, treatment, or resolution. Therefore, the chronicity of an individual’s disease is unknown. Furthermore, although the time course for many of the commonly discussed diseases is typically over months and years rather than days or weeks, disease severity or certain treatments can produce transient changes in liver function, kidney function, and blood sugar regulation, to name a few [10, 47].

Additionally, temporal relationships between cancer diagnosis and development of other diseases were difficult to assess due to the lack of information regarding age at diagnosis, severity, and resolution of diseases. Cancer-specific temporal information—including onset of chemotherapy or radiation, severity/staging of cancer, remission, recurrence, and metastasis—is important to understanding the association between multimorbidity and cancer diagnosis [3]. Some of these variables (metastasis, current treatment, and recurrence) were included in the Cancer Supplement in previous years but were not included in the 2018 NHIS survey. Our study excluded: survivors of childhood cancer (defined as those diagnosed with cancer before age of 21); adults reporting multiple cancers; or adults diagnosed with a cancer other than those identified for this analysis. Additional studies can examine multimorbidity clusters in these groups, as different patterns may emerge.

Our study aimed to include several demographic characteristics, including the “oldest old” population and BMI. However, in NHIS, all adults over the age of 85 are coded as 85 (https://​ftp.​cdc.​gov/​pub/​Health_​Statistics/​NCHS/​Dataset_​Documentation/​NHIS/​2018/​samadult_​layout.​pdf). Therefore, age cannot be used as a continuous variable above the age of 85, which is the age group with the most rapidly increasing incidence of cancer [47]. Additionally, NHIS does not sample institutionalized individuals, such as those in nursing home or other long-term care facilities, which disproportionately impacts the “oldest old” population in the United States. Similarly, BMI was not calculated for all participants because the lowest and highest heights and weights are considered extreme categories and are not included in the data set (https://​ftp.​cdc.​gov/​pub/​Health_​Statistics/​NCHS/​Dataset_​Documentation/​NHIS/​2018/​samadult_​layout.​pdf). This limitation impacted our ability to provide insight on the oldest old (85 +) population, which is often absent in existing literature. Furthermore, it limited our ability to discuss cachectic and underweight older adults, which can often impact their health, especially among those with cancer [48].

Additional variables associated with multimorbidity clusters in adults older than 65 with and without cancer, such as HRQOL, were not included in this study. The Patient-Reported Outcomes Measurement Information System (PROMIS) questionnaire (https://​www.​healthmeasures.​net/​explore-measurement-systems/​promis), which measures HRQOL, was previously administered in the Cancer Supplement of NHIS but was not included the 2018 survey. These limitations demonstrate that existing data sources were not created for this type of complex research; thus, it may contribute to the scarcity of cluster analysis and multimorbidity research, especially at a population level.

Finally, our study aimed to understand the differences in disease clusters among adults with cancer and controls. As one of the first studies to use NHIS data for a disease cluster analysis not in relation to a primary diagnosis or within a subpopulation of the US, it was important to analyze diseases in relationship to cancer status, not geographic or demographic distribution. Therefore, the analysis was conducted using data without survey weights. Thus, our results are not generalizable, or nationally representative, and do not account for biases incurred in the sampling process, such as non-response and social desirability bias. Additionally, recall period bias may lead to underreporting of diseases diagnosed at younger ages and overreporting of diseases diagnosed at older ages, especially when focused on self-report data from older adults in a cross-sectional study. Future studies can compare self-report data to other sources of data, such as electronic medical record or claims data, to investigate the impact of recall period bias on self-report of diseases in surveys such as NHIS. However, this was out of the scope of our research.