This was a pilot prospective longitudinal cohort study of 50 adults ≥60 years of age with any stage of untreated or recurrent NSCLC. Eligible participants underwent a baseline assessment using the CARG CGA (Hurria et al., 2005; Nie et al., 2013). Briefly, this assessment included the CARG chemotherapy toxicity calculator, the Patient-Reported Outcome Measurement Information System (PROMIS 10 (Hays et al., 2009) Global Health Scale Short Form v1.1, and sociodemographic information. The CARG CGA includes self-reported functional status measures, comorbidity, medications, nutrition, psychological state, and social support. A clinical research coordinator administered the cognitive assessment, the Blessed Orientation-Memory-Concentration test (BOMC test (Tuch et al., 2021)), and the short physical performance battery (SPPB) (Guralnik et al., 1994; Kawas et al., 1995). Participants also underwent longitudinal peripheral blood draws and stool sample collection. Participants then received standard-of-care lung cancer treatment or participated in a clinical trial as decided by their treating physician.

Participants were recruited, consented, and enrolled in the FITNESS study at The Ohio State University Comprehensive Cancer Center (OSUCCC)–James Thoracic Oncology clinic. Eligible patients were ≥60 years of age and intended to receive systemic treatment at the OSUCCC for either a new or recurrent NSCLC diagnosis. Participants were required to have decision-making capacity and willingness to sign informed consent. Patients with any medical conditions, including but not limited to unstable angina pectoris, cardiac arrhythmia, psychiatric illness, or cognitive impairment, without a legally authorized representative (LAR) that would verify compliance with study procedures were excluded.

Participants underwent a baseline assessment before initiating lung cancer treatment. This process included sociodemographic and disease assessment; CARG toxicity tool calculation (Hurria et al., 2011); stool and blood sample collection; and completion of the CARG CGA, the BOMC, the SPPB (Almugbel et al., 2022), and the timed up-and-go (TUG) (Nakano et al., 2021).

The type of treatment and drug doses (planned and actual) were recorded along with any supportive care medications given throughout treatment. Grade 1–5 toxicities (National Cancer Institute Common Terminology Criteria for Adverse Events, NCI CTCAE V5) (Kluetz et al., 2016), hospitalizations, dose delays and reductions, and drug discontinuation were captured. The toxicity assessment and chemotherapy drug and dose information were collected by a study coordinator at the beginning of every treatment cycle, reviewed and attributed by two certified oncologists. Vital signs, labs, and health-related quality of life questionnaires, including the PROMIS 10²⁸ and European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire for Lung Cancer (QLQ-LC-13) (Bergman et al., 1994), were collected at visit 2 (day 90) and visit 3 (day 180). Patients underwent additional monthly functional status assessments (FSA), monthly biospecimen banking, and an additional CGA at visit 3.

Baseline assessments were repeated at the end-of-treatment (EOT) visit. These occurred up to 45 days after completing treatment or 365 days after treatment initiation—whichever occurred first. Due to the varying length of treatment plans and different staging at the time of diagnosis, participants prescribed only four cycles of chemotherapy (adjuvant treatment) were only eligible to participate in the baseline and EOT visits. Visit 1 (baseline), visit 2 (day 90), and visit 3 (day 180) allowed a ±30-day window of time, letting patients enroll up to 30 days after treatment initiation. Disease response was calculated using Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 criteria by two medical oncology fellows using chest cat scan (CT) imaging (Eisenhauer et al., 2009). Target lesions were followed throughout the study at each visit in which imaging was available. Tumor response was categorized into four groups: complete response, partial response, stable disease, and progressive disease.

Stool samples were collected at baseline and during participants’ oncology visits, roughly every 3–4 weeks. Patients were provided with a stool collection tub and collection instructions. Patients were asked to collect a sample within 24 h of returning to the clinic. Once the biospecimen sample reached the lab, it was aliquoted into cryovials and stored at −80°C. Samples were then processed in batches. DNA was extracted using the QIAGEN Power Fecal Pro kit. Sequencing was performed on the Illumina NovaSeq 6,000 platform to a depth of eight million reads per sample. Demultiplexed fastq files were cleaned, paired, and taxonomically assigned following the MetaPhlAn HUMAnN 3 Pipeline (Beghini et al., 2021).

Peripheral blood (40 mL) was collected at baseline, 6 months, and at either 1 year or the end of the patient’s treatment plan. Blood samples were only collected from participants who had not initiated treatment before visit one to control the purity of data. The samples were processed by The Ohio State University Leukemia Tissue Bank (LTB). Peripheral blood T cells were isolated via negative selection using the RosetteSep human T cell isolation cocktail (Stemcell Technologies). Tcell RNA was isolated using a Zymo Research quick-DNA/RNA miniprep kit. The quality and concentration of the resulting RNA was quantified on an Agilent TapeStation and analyzed using a custom NanoString codeset panel, OSU Senescence. The OSU Senescence panel is comprised of 74- T-cell markers with five loading controls. Gene expression markers were selected to measure subset-specific T-cell mRNA markers of cellular senescence and T-cell differentiation proliferation, and exhaustion as previously published (Rosko et al., 2019; Lucas et al., 2020; Fell et al., 2022). Additionally, certain transcription profiles in T cells such as LAG3 and CD8 transcription expression has been shown to be associated with inferior clinical outcomes in patients with myeloma (Lucas et al., 2020). T cells are exhausted when cells lose cytotoxic capacity, show sustained expression of inhibitory checkpoint molecules such as LAG3 and have distinct altered transcription profiles in CD8 subsets (Lucas et al., 2020). Therefore, LAG3 and CD8 transcription profiles were examined for associations with longitudinal IADL scores.

The primary aim of this study was to assess the feasibility of a longitudinal geriatric assessment collected concurrently with blood and stool biomarkers. Metrics included the number of blood and stool samples collected, the completion of PRO assessments, and patient withdrawal. Feasibility was defined a priori as the proportion of missing information at each time point, with success defined as at least 50% of participants able to complete study visits 2, 3, and four or EOT. Reasons for missed collections and study withdrawals were captured.

Descriptive statistics were used to summarize demographic information, with totals and percentages used for categorical data and medians and interquartile ratio (IQR) used for continuous data. When determining the immune-related adverse event (irAE) grade for a given time point, the grade of the most extreme irAE for that time point was used. For association with baseline CARG, the maximum-grade irAE experienced by the patient was considered.

Microbe and gene expression associations with the CGA were performed using longitudinal mixed-effects models using the {lme4} and {lmerTest} packages in R (Bates et al., 2015). A model was run for each combination of GA tool (i.e., SPPB or IADL), microbe, and gene. For associations between microbes and the selected GA tool, each GA tool was considered an independent variable and all other variables were considered dependent variables. Repeated measures in patients were handled using random effects for each patient. All other effects were fixed. The longitudinal nature of the data was accounted for by including a term for the number of days on the study associated with each measure. The models can be described by the following equation, in which the term A _i is the score (A) for each geriatric assessment(i), M _j is the relative abundance (M) for each microbe (j), D is days on study, C is immune checkpoint inhibitor, (1|P _k ) is the intercept (1|P) for each patient (k): A i , k = M j , k + D + C + 1 | P k

For transcript and geriatric assessment tools, variables were centered and scaled before modeling. Geriatric assessments remained the independent variable and patient intercepts remain random effects. The effects of change over time are still controlled for by including days on study in the model. These models can be described with the following equation, where A _i is the score (A) geriatric assessment of interest (i), G _j is the expression (G) of the gene of interest (i), D is days on study, (1|P _k ) is the intercept (1|P) for each patient (k): A i , k = G j , k + D + 1 | P k

Analyses were performed in R, version 4.1.

The study enrolled 61 patients with 11 withdrawals prior to starting (18% dropout rate), resulting in 50 evaluable participants at baseline. Among the 11 patients who dropped out before the study began, the reasons for withdrawal included disease progression, declining performance status, unwillingness to complete the study assessments, and study burden, including increased visit time and survey fatigue. Among the 50 participants, one withdrew at visit one due to cognitive impairment without an LAR. Reasons for drop out prior to the 3-month time point were death (n = 3) and withdrawal (n = 4). Inclusion criteria required participants to be undergoing active treatment versus surveillance for longitudinal follow up; therefore, the three- and 6-month time points excluded an additional 11 participants as they had completed treatment (adjuvant chemotherapy). Excluding the 11 adjuvant chemotherapy participants, at 3-months, 28 of 31 eligible participants (90%) completed the visit two assessments. Excluding the 11 adjuvant chemotherapy patients, at 6 months, 23 of 25 eligible participants (92%) had completed the study assessments (study visit 3). Reasons for drop out prior to 6 months were death (n = 4) and withdrawal (n = 2). For study visit 4, 30 (97%) of the remaining 31 participants completed the assessments. Reasons for drop out prior to 1-year were death (n = 4) and withdrawal (n = 1). The overall study attrition rate was 38%, excluding those who wished to continue participation after refusing to complete at least one study visit. Common reasons for study drop out included death (22%), loss of interest in participation (14%), and cognitive impairment (2%)—The study CONSORT diagram in Figure 1A details study enrollment, follow-up, and dropout.

The collection of longitudinal gut microbiome samples posed challenges, including patient hesitancy, transportation difficulties for those with longer travel times, and constipation before the return-to-clinic visit. Despite these barriers, 113 stool samples were collected, including 24 baseline (treatment-naive) specimens. Of those, 23 (95.8%) participants provided longitudinal samples for correlative analysis. See Figure 1B for a detailed overview of longitudinal biomarker collection.

The mean age at baseline was 71.7 years (60–88 years). Most participants were Caucasian (92%), and 58% of patients were male. The most common histology was adenocarcinoma (70%), followed by squamous cell carcinoma (24%). Most cases were advanced stage (90% stage III/IV). For first-line treatment, 44% of patients received chemotherapy, 30% received a combination of chemotherapy and ICIs, 22% received ICIs alone, and 2 (4%) received TKIs. The mean CARG chemotherapy toxicity score at baseline was 7.5 (range 2–12), indicating an intermediate risk of treatment toxicity. Patient demographics and characteristics are further detailed in Table 1. Patient treatment timelines with the treatment types, treatment responses, and events, including toxicities and biomarker obtainment, are displayed in Figure 2.

The CARG-CGA subset on IADLs (median: 14), Medical Outcomes Study (MOS) Physical Health (median: 61), Self-report Health Rating (median: 90), and the MOS Mental Health Inventory (median: 81) indicate relatively high levels of independence and mental health statuses among this cohort at baseline. Similarly, the PROMIS 10 average functional disability score (median: 24) and the EORTC-LC subset score (median: 17) indicated minimal to no baseline disability and minimal levels of symptom burden (Table 2; complete results including longitudinal data appear in Supplementary Table S4). While participants scored highly on the Social Support Medical Outcomes Survey (median: 96), their scores demonstrated a moderate impact to their usual levels of social activities (median: 58). The CARG-CGA Comorbidity Index showed that participants had an average of six pre-existing comorbid conditions present at baseline.

Using the CARG chemotherapy toxicity tool, 22 patients (47%) had a low or mild risk of chemotherapy toxicity, and 25 (53%) had a high risk of chemotherapy toxicity ( Table 2). CARG toxicity scores were useful in predicting the presence of at least one grade 3 or higher adverse event (Figure 3). The majority (74%) of participants experienced a grade 1 or two treatment-related adverse events and 10 (20%) of participants experienced a grade 3 or higher adverse event. Participants who scored ≤7 on the CARG toxicity assessment tool (47%) had three participants with a ≥grade 3 while those that scored ≥8 (53%) included seven participants that experienced a ≥grade 3 (70%) that occurred during this study. Patient-reported outcomes measuring physical function (IADLs, PROMIS 10, MOS) among this cohort did not show high levels of disability while the SPPB suggests the presence of physical impairment (median: 9) and 19 participants (38%) had TUG scores that indicated a mild risk for falls. At baseline, participants showed a mild to moderate concern for cognitive impairment (BOMC (Tuch et al., 2021) median: 6) (Table 2).

Changes in GA tools over time were associated with changes in molecular features, including the gut microbiome and T-cell senescence markers. For example, Lactobacillus rogosae, was associated with increased (improving) IADL, QoL, and SPPB scores (Figure 4A; complete results with all p-values corrected for multiple hypothesis testing appear in Supplementary Table S1). In contrast, the microbes Romboutsia ilealis, Streptococcus, and Lachnoclostridium sp An138 were associated with decreased (worsening) IADL, QoL, and SPPB scores over time (Figure 4A). Similarly, T-cell lag3 and cd8a RNA expression over time were associated with decreased IADL scores (Figure 4B; complete results with all p-values corrected for multiple hypothesis testing appear in Supplementary Table S2, with results of an additional sensitivity analysis of these results using Spearman correlations appearing in Supplementary Table S3 and the full dataset appearing in Supplementary Table S4).