This study utilized data from the NHIS spanning the years 1997 to 2018. The NHIS is a nationally representative, cross-sectional survey conducted by the National Center for Health Statistics (NCHS) to assess health status, behaviors, and sociodemographic factors among the U.S. population.
A total of 671,696 participants were initially identified from the NHIS 1997-2018 dataset. Sequential exclusion criteria were applied to ensure data completeness and study relevance. Participants with missing data on frailty assessment (N = 28,461) were excluded, followed by those with missing information on key covariates (N = 47,238). Individuals with a history of cancer other than lung cancer (N = 467) were also excluded. Additionally, participants with missing mortality outcome data (N = 5742) were removed.
After these exclusions, 589,788 individuals remained eligible for analysis. Among them, 1,778 were identified as lung cancer survivors, while 588,010 had no history of cancer. The final analytic cohort consisted of these two groups, allowing for the examination of frailty status in relation to all-cause mortality (Fig. 1).
Frailty was assessed using a modified FRAIL scale, a validated screening instrument originally developed by the Geriatric Advisory Panel of the International Society for Nutrition and Aging. For this study, we derived frailty measures from the 1997-2018 NHIS, utilizing self-reported questionnaire responses. The FRAIL scale consists of five domains: Fatigue, Resistance, Ambulation, Illness, and Low body mass index (BMI).
Fatigue was assessed using NHIS survey questions regarding the frequency of experiencing tiredness or low energy over a predefined recall period. Responses indicating frequent or persistent fatigue were scored as 1, whereas those reflecting minimal or no fatigue were scored as 0.
Resistance was evaluated based on self-reported difficulty in ascending or descending 12 steps without assistance or assistive devices. Ambulation was assessed by asking whether the participant experienced difficulty walking 100 yards on level ground (approximately the length of a football field or a city block) unaided. Participants reporting no difficulty in either task were assigned a score of 0, while those reporting any degree of difficulty received a score of 1.
The illness component was derived from self-reported physician-diagnosed chronic conditions. Participants reporting five or more conditions from a predefined list of 12 diseases — angina, anxiety disorder, arthritis, asthma, chronic obstructive pulmonary disease, coronary heart disease, dementia, diabetes, myocardial infarction, hyperlipidemia, hypertension, and stroke — were assigned a score of 1, while those reporting fewer than five were assigned a score of 0.
Low BMI was defined as a body mass index (BMI) below 18.5 kg/m, with affected individuals receiving a score of 1 and all others assigned a score of 0.
The total FRAIL scale score ranged from 0 to 5, with participants categorized as frail (scores of 3-5), pre-frail (scores of 1-2), or robust (score of 0).
In this study, we utilized data from the NHIS to assess various covariates, including age, sex, race/ethnicity, education level, health insurance status, marital status, geographic region, and depression status, time since cancer diagnosis, number of cancer diagnoses, and number of comorbidities. The NHIS is a nationally representative, cross-sectional survey conducted annually by the NCHS to monitor the health of the civilian non-institutionalized U.S. population.
Participants reported their age in years at the time of the interview and self-identified their sex as either male or female. These demographic variables are standard components of the NHIS and are collected through direct questioning during household interviews. ace and Hispanic origin were determined based on self-identification. Participants were asked to select one or more races that they considered themselves to be, and to indicate whether they were of Hispanic or Latino origin. For analytical purposes, race/ethnicity was categorized into White, Black, Asian, and Other. Educational attainment was assessed by inquiring about the highest level of school completed or the highest degree received. Responses were categorized as less than high school, high school graduate, and more than high school. Participants were asked about their health insurance coverage at the time of the interview, including private health insurance, Medicare, Medicaid, or other government-sponsored health plans. Responses were dichotomized into insured (yes) or uninsured (no). Marital status was determined by asking participants to describe their current marital situation, with responses categorized as married or unmarried. The NHIS classifies participants’ residence into four geographic regions: Northeast, Midwest, South, and West, based on the U.S. Census Bureau’s regional definitions. Depression was assessed through self-reported data, where participants indicated whether they had ever been told by a healthcare professional that they had depression. This method of assessment aligns with approaches used in prior studies analyzing NHIS data. By employing these standardized NHIS measures, we ensured the reliability and validity of the covariate data utilized in our analyses.
Information on cancer history was also based on self-reported responses in the NHIS. Participants who responded “yes” to the question “Have you ever been told by a doctor or other health professional that you had cancer or a malignancy of any kind?” were classified as having a history of cancer. Time since cancer diagnosis was calculated as the difference between the year of interview and the self-reported year of the first cancer diagnosis, and was categorized as < 2 years or ≥ 2 years. The number of cancer diagnoses was derived from the total number of different cancer types reported by each participant, and was classified as 1 or ≥ 2. Both variables were included as categorical covariates in multivariable analyses to account for the potential confounding effect of cancer history. The number of comorbidities was defined as the total count of self-reported physician-diagnosed chronic conditions (angina, anxiety disorder, arthritis, asthma, chronic obstructive pulmonary disease, coronary heart disease, dementia, diabetes, myocardial infarction, hyperlipidemia, hypertension, and stroke) and categorized for analysis as 0-1, 2-3, or ≥ 4.
To facilitate mortality follow-up, the NHIS data from 1997 to 2018 were linked to mortality outcomes through the NDI, enabling comprehensive mortality follow-up through December 31, 2019. This linkage was facilitated by the NCHS Data Linkage Program, which employs a probabilistic matching methodology to align NHIS records with NDI death certificate data. This process allows for the assessment of vital status and cause-specific mortality among survey participants.
The linkage process adhered to stringent confidentiality protocols to protect respondent privacy. In the public-use Linked Mortality Files (LMF), data perturbation techniques were applied to reduce the risk of participant re-identification. For select records, synthetic data were substituted for follow-up time or underlying cause of death; however, information regarding vital status remained unaltered.
Utilizing the NHIS Linked Mortality Files, we conducted a comprehensive analysis of mortality outcomes among study participants. This approach allowed us to investigate the associations between frailty status and all-cause mortality, leveraging the rich health and demographic data collected by the NHIS.
All statistical analyses were conducted using SAS software, version 9.4 (Cary, North Carolina, USA) and R software (version 4.3.1; https://www.R-project.org). Differences in categorical variables between groups were assessed using the chi-square (χ) test, while differences in continuous variables were evaluated using one-way analysis of variance (ANOVA).
Survival analyses were conducted using Kaplan-Meier curves to estimate survival probabilities, with comparisons between groups assessed using the log-rank test. Multivariable Cox proportional hazards regression models were employed to evaluate the association between frailty status and all-cause mortality, adjusting for potential confounders. The proportional hazards assumption was assessed using Schoenfeld residuals.
All statistical tests were two-sided, and a P-value of less than 0.05 was considered to indicate statistical significance.
