Continuous Glucose Monitoring in Older Adults: What We Know and What We Have Yet to Learn.

OBJECTIVE: To assess the growing use of continuous glucose monitoring (CGM) systems by older adults and explore additional areas integration that could benefit adults with frailty. BACKGROUND: The use of CGM devices has expanded rapidly in the last decade. This has been supported by substantial data showing significant benefit in glycemic metrics: hemoglobin A1c improvements, less hypoglycemia, and improved quality of life. However, sub-populations, such as older persons, exist where available data are limited. Furthermore, frail older adults represent a heterogeneous population with their own unique challenges to the management of diabetes. This group has some of the poorest outcomes related to the sequela of diabetes. For example, hypoglycemia resulting in significant morbidity and mortality is more frequent in older person with diabetes than in younger persons with diabetes. METHOD: We present a concise literature review on CGM use in the older adult as well as expand upon glycemic and nonglycemic benefits of CGM for patients, caregivers, and providers. Retrospective analysis of inpatient glycemic data of 16,935 older adults with Type 2 diabetes mellitus at Atrium Health Wake Forest Baptist indicated those with fraility managed with insulin or sulfonylurea had the highest rates of delirium (4.8%), hypoglycemia (3.5%), cardiovascular complications (20.2%) and ED visits/hospitalizatoins (49%). In addition, we address special consideration of specific situations including inpatient, palliative and long term care settings. CONCLUSION: This review article summarizes the available data for CGM use in older adults, discusses the benefits and obstacles with CGM use in this population, and identifies areas of future research needed for improved delivery of care to older persons with diabetes.

New horizons in frailty identification via electronic frailty indices: early implementation lessons from experiences in England and the United States.

Frailty represents an integrative prognostic marker of risk that associates with a myriad of age-related adverse outcomes in older adults. As a concept, frailty can help to target scarce resources and identify subgroups of vulnerable older adults that may benefit from interventions or changes in medical management, such as pursing less aggressive glycaemic targets for frail older adults with diabetes. In practice, however, there are several operational challenges to implementing frailty screening outside the confines of geriatric medicine. Electronic frailty indices (eFIs) based on the theory of deficit accumulation, derived from routine data housed in the electronic health record, have emerged as a rapid, feasible and valid approach to screen for frailty at scale. The goal of this paper is to describe the early experience of three diverse groups in developing, implementing and adopting eFIs (The English National Health Service, US Department of Veterans Affairs and Atrium Health-Wake Forest Baptist). These groups span different countries and organisational complexity, using eFIs for both research and clinical care, and represent different levels of progress with clinical implementation. Using an implementation science framework, we describe common elements of successful implementation in these settings and set an agenda for future research and expansion of eFI-informed initiatives.

The Association of Frailty and Neighborhood Disadvantage with Emergency Department Visits and Hospitalizations in Older Adults.

BACKGROUND: Risk stratification and population management strategies are critical for providing effective and equitable care for the growing population of older adults in the USA. Both frailty and neighborhood disadvantage are constructs that independently identify populations with higher healthcare utilization and risk of adverse outcomes. OBJECTIVE: To examine the joint association of these factors on acute healthcare utilization using two pragmatic measures based on structured data available in the electronic health record (EHR). DESIGN: In this retrospective observational study, we used EHR data to identify patients aged ≥ 65 years at Atrium Health Wake Forest Baptist on January 1, 2019, who were attributed to affiliated Accountable Care Organizations. Frailty was categorized through an EHR-derived electronic Frailty Index (eFI), while neighborhood disadvantage was quantified through linkage to the area deprivation index (ADI). We used a recurrent time-to-event model within a Cox proportional hazards framework to examine the joint association of eFI and ADI categories with healthcare utilization comprising emergency visits, observation stays, and inpatient hospitalizations over one year of follow-up. KEY RESULTS: We identified a cohort of 47,566 older adults (median age = 73, 60% female, 12% Black). There was an interaction between frailty and area disadvantage (P = 0.023). Each factor was associated with utilization across categories of the other. The magnitude of frailty's association was larger than living in a disadvantaged area. The highest-risk group comprised frail adults living in areas of high disadvantage (HR 3.23, 95% CI 2.99-3.49; P < 0.001). We observed additive effects between frailty and living in areas of mid- (RERI 0.29; 95% CI 0.13-0.45; P < 0.001) and high (RERI 0.62, 95% CI 0.41-0.83; P < 0.001) neighborhood disadvantage. CONCLUSIONS: Considering both frailty and neighborhood disadvantage may assist healthcare organizations in effectively risk-stratifying vulnerable older adults and informing population management strategies. These constructs can be readily assessed at-scale using routinely collected structured EHR data.

The Association Between an Electronic Health Record (EHR)-Embedded Frailty Index and Patient-Reported Outcomes Among Patients with Metastatic Non-Small-Cell Lung Cancer on Immunotherapy: A Brief Report.

Background: While frailty is a well-established predictor of overall mortality among patients with metastatic non-small cell lung cancer (mNSCLC), its association with patient-reported outcomes is not well-characterized. The goal of this study was to examine the association between an electronic frailty index (eFI) score and patient-reported outcome measures along with prognostic awareness among patients with mNSCLC receiving immunotherapy. Methods: In a cross-sectional study, patients with mNSCLC who were on immunotherapy completed the European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire Core 30 (EORTC-QLQ-C30) and the National Cancer Institute Patient Reported Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE). We utilized bivariate analyses to compare quality of life, symptoms, supportive services, and prognostic awareness among 3 groups defined by e-frailty status. Results: Sixty patients (mean age 62.5 years, 75% Caucasian, 60% women) participated. Most patients were pre-frail (68%), with 13% being frail and 18% non-frail. Pre-frail and frail patients had significantly lower physical function scores (mean 83.9 fit vs 74.8 pre-frail vs 60.0 frail, P = .04) and higher rates of self-reported pain (75% frail vs 41.5% pre-frail vs 18.2% fit; P = .04) compared to non-frail patients. We found no differences in palliative referral rates. Conclusion: Pre-frail and frail mNSCLC patients identified by the eFI have higher rates of pain and physical functional impairments than non-frail patients. These findings highlight the importance of emphasizing preventive interventions targeting social needs, functional limitations, and pain management, especially among pre-frail patients to reduce further decline.

Assessment of an embedded primary care-derived electronic health record (EHR) frailty index (eFI) in older adults with acute myeloid leukemia.

INTRODUCTION: Assessing frailty is integral to treatment decision-making for older adults with acute myeloid leukemia (AML). Prior electronic frailty indices (eFI) derive from an accumulated-deficit model and are associated with mortality in older primary care populations. We evaluated use of an embedded eFI in AML by describing baseline eFI categories by treatment type and exploring associations between eFI categories, survival, and treatment received. MATERIALS AND METHODS: This was a retrospective study of subjects ≥60 years old with new AML treated at an academic medical center from 1/2018-10/2020. The eFI requires ≥2 ambulatory visits over two years and uses demographics, vitals, ICD-10 codes, outpatient labs, and available functional information from Medicare Annual Wellness Visits. Frailty was defined as fit (eFI ≤ 0.10), pre-frail (0.10 < eFI ≤ 0.21), and frail (eFI > 0.21). Chemotherapy was intensive (anthracycline-based) or less-intensive (hypomethylating agent, low dose cytarabine +/- venetoclax). Therapy type, pre-treatment characteristics, and chemotherapy cycles were compared by eFI category using chi-square and Fisher's exact tests and ANOVA. Median survival was compared by eFI category using log-rank tests stratified by therapy type. RESULTS: Among 166 older adults treated for AML (mean age 74 years, 61% male, 85% Caucasian), only 79 (48%) had a calculable eFI score before treatment. Of these, baseline eFI category was associated with treatment received (fit (n = 31): 68% intensive, 32% less intensive; pre-frail (n = 38): 37% intensive, 63% less intensive; frail (n = 10): 0% intensive, 100% less intensive; not calculable (n = 87): 48% intensive, 52% less-intensive; p < 0.01). The prevalence of congestive heart failure and secondary AML differed by frailty status (p < 0.01). Median survival did not differ between eFI categories for intensively (p = 0.48) or less-intensively (p = 0.09) treated patients. For those with less-intensive therapy who lived ≥6 months, eFI category was not associated with the number of chemotherapy cycles received (p = 0.97). The main reason for an incalculable eFI was a lack of outpatient visits in our health system prior to AML diagnosis. DISCUSSION: A primary care-derived eFI was incalculable for half of older adults with AML at an academic medical center. Frailty was associated with chemotherapy intensity but not survival or treatment duration. Next steps include testing adaptations of the eFI to the AML setting.

ACR Appropriateness Criteria® Headache: 2022 Update.

Headache is an ancient problem plaguing a large proportion of the population. At present, headache disorders rank third among the global causes of disability, accounting for over $78 billion per year in direct and indirect costs in the United States. Given the prevalence of headache and the wide range of possible etiologies, the goal of this document is to help clarify the most appropriate initial imaging guidelines for headache for eight clinical scenarios/variants, which range from acute onset, life-threatening etiologies to chronic benign scenarios. The American College of Radiology Appropriateness Criteria are evidence-based guidelines for specific clinical conditions that are reviewed annually by a multidisciplinary expert panel. The guideline development and revision process support the systematic analysis of the medical literature from peer reviewed journals. Established methodology principles such as Grading of Recommendations Assessment, Development, and Evaluation or GRADE are adapted to evaluate the evidence. The RAND/UCLA Appropriateness Method User Manual provides the methodology to determine the appropriateness of imaging and treatment procedures for specific clinical scenarios. In those instances where peer reviewed literature is lacking or equivocal, experts may be the primary evidentiary source available to formulate a recommendation.

Development of an Advance Care Planning Portal-Based Tool for Community-Dwelling Persons Living With Cognitive Impairment: The ACPVoice Tool.

Background: Patient portals can be an innovative and efficient way to engage patients in advance care planning (ACP). However, comprehension and judgment in older adults with cognitive impairment presents several barriers and challenges to engaging in new technology. Our objective was to develop an ACP portal-based tool (ACPVoice) for community-dwelling persons living with cognitive impairment (PLCI) by engaging end-users in the design process. Methods: Two rounds of cognitive interviews were conducted to identify and resolve cognitive issues related to comprehension, judgment, response, and to assess content validity. Purposive sampling was used with the goal of enrolling 15 different participants (five with mild cognitive impairment and five dyads (those with mild dementia and their care partner) in each round to assess respondents' understanding of questions related to advance care planning to be administered via the patient portal. Results: Twenty PLCI (mean age 78.4, 10 females [50%]) and ten care partners (mean age 60.9, 9 females [90%]) completed cognitive interviews between May 2021 and October 2021. The mean Mini-Mental State Examination score for PLCI was 25.6 (SD 2.6). Unclear wording and undefined vague and/or unfamiliar terms were the major issues identified. Revisions to item wording, response options, and instructions were made to improve question comprehension and response as well as navigational ease. Conclusion: Minor changes to the wording, format, and response options substantially improved respondents' ability to interpret the item content of the ACPVoice tool. Dissemination and implementation of the ACPVoice tool could help to engage community-dwelling PLCI in ACP discussions.

Automated Electronic Frailty Index is Associated with Non-home Discharge in Patients Undergoing Open Revascularization for Peripheral Vascular Disease.

BACKGROUND: Frailty is associated with adverse surgical outcomes including post-operative complications, needs for post-acute care, and mortality. While multiple frailty screening tools exist, most are time and resource intensive. Here we examine the association of an automated electronic frailty index (eFI), derived from routine data in the Electronic Health Record (EHR), with outcomes in vascular surgery patients undergoing open, lower extremity revascularization. METHODS: A retrospective analysis at a single academic medical center from 2015 to 2019 was completed. Information extracted from the EHR included demographics, eFI, comorbidity, and procedure type. Frailty status was defined as fit (eFI≤0.10), pre-frail (0.100.21). Outcomes included length of stay (LOS), 30-day readmission, and non-home discharge. RESULTS: We included 295 patients (mean age 65.9 years; 31% female), with the majority classified as pre-frail (57%) or frail (32%). Frail patients exhibited a higher degree of comorbidity and were more likely to be classified as American Society of Anesthesiologist class IV (frail: 46%, pre-frail: 27%, and fit: 18%, P = 0.0012). There were no statistically significant differences in procedure type, LOS, or 30-day readmissions based on eFI. Frail patients were more likely to expire in the hospital or be discharged to an acute care facility (31%) compared to pre-frail (14%) and fit patients (15%, P = 0.002). Adjusting for comorbidity, risk of non-home discharge was higher comparing frail to pre-frail patients (OR 3.01, 95% CI 1.40-6.48). DISCUSSION: Frail patients, based on eFI, undergoing elective, open, lower extremity revascularization were twice as likely to not be discharged home.

Using an Electronic Health Record and Deficit Accumulation to Pragmatically Identify Candidates for Optimal Prescribing in Patients With Type 2 Diabetes.

Objective: Despite guidelines recommending less stringent glycemic goals for older adults with type 2 diabetes, overtreatment is prevalent. Pragmatic approaches for prioritizing patients for optimal prescribing are lacking. We describe glycemic control and medication patterns for older adults with type 2 diabetes in a contemporary cohort, exploring variability by frailty status. Research Design and Methods: This was a cross-sectional observational study based on electronic health record (EHR) data, within an accountable care organization (ACO) affiliated with an academic medical center/health system. Participants were ACO-enrolled adults with type 2 diabetes who were ≥65 years of age as of 1 November 2020. Frailty status was determined by an automated EHR-based frailty index (eFI). Diabetes management was described by the most recent A1C in the past 2 years and use of higher-risk medications (insulin and/or sulfonylurea). Results: Among 16,973 older adults with type 2 diabetes (mean age 75.2 years, 9,154 women [53.9%], 77.8% White), 9,134 (53.8%) and 6,218 (36.6%) were classified as pre-frail (0.10 < eFI ≤0.21) or frail (eFI >0.21), respectively. The median A1C level was 6.7% (50 mmol/mol) with an interquartile range of 6.2-7.5%, and 74.1 and 38.3% of patients had an A1C <7.5% (58 mmol/mol) and <6.5% (48 mmol/mol), respectively. Frailty status was not associated with level of glycemic control (P = 0.08). A majority of frail patients had an A1C <7.5% (58 mmol/mol) (n = 4,544, 73.1%), and among these patients, 1,755 (38.6%) were taking insulin and/or a sulfonylurea. Conclusion: Treatment with insulin and/or a sulfonylurea to an A1C levels <7.5% is common in frail older adults. Tools such as the eFI may offer a scalable approach to targeting optimal prescribing interventions.

Frailty and COVID-19 mRNA Vaccine Antibody Response in the COVID-19 Community Research Partnership.

BACKGROUND: COVID-19 has disproportionately affected older adults. Frailty has been associated with impaired vaccine response in other vaccine types, but the impact of frailty on mRNA vaccine response is undefined. METHODS: Observational study of adults aged 55 and older from 1 U.S. health care system between January 22, 2021 and September 16, 2021 with self-reported Moderna or Pfizer COVID-19 mRNA vaccine and an electronic frailty index (eFI) score from their medical record (n = 1 677). Participants' frailty status was compared with positive antibody detection (seroconversion) following full vaccination and subsequent loss of positive antibody detection (seroreversion) using logistic regression models. RESULTS: Of 1 677 older adults with median (interquartile range) age, 67 (62 and 72) years, and frailty status (nonfrail: 879 [52%], prefrail: 678 [40%], and frail: 120 [7.2%]), seroconversion was not detected in 23 (1.4%) over 60 days following full vaccination. Frail individuals were less likely to seroconvert than nonfrail individuals, adjusted odds ratio (OR) 3.75, 95% confidence interval (CI; 1.04, 13.5). Seroreversion was detected in 50/1 631 individuals (3.1%) over 6 months of median follow-up antibody testing. Frail individuals were more likely to serorevert than nonfrail individuals, adjusted OR 3.02, 95% CI (1.17, 7.33). CONCLUSION: Overall antibody response to COVID-19 mRNA vaccination was high across age and frailty categories. While antibody detection is an incomplete descriptor of vaccine response, the high sensitivity of this antibody combined with health-system data reinforce our conclusions that frailty is an independent predictor of impaired antibody response to the COVID-19 mRNA vaccines. Frailty should be considered in vaccine studies and prevention strategies.

Predicting Future Mobility Limitation in Older Adults: A Machine Learning Analysis of Health ABC Study Data.

BACKGROUND: Mobility limitation in older adults is common and associated with poor health outcomes and loss of independence. Identification of at-risk individuals remains challenging because of time-consuming clinical assessments and limitations of statistical models for dynamic outcomes over time. Therefore, we aimed to develop machine learning models for predicting future mobility limitation in older adults using repeated measures data. METHODS: We used annual assessments over 9 years of follow-up from the Health, Aging, and Body Composition study to model mobility limitation, defined as self-report of any difficulty walking a quarter mile or climbing 10 steps. We considered 46 predictors, including demographics, lifestyle, chronic conditions, and physical function. With a split sample approach, we developed mixed models (generalized linear and Binary Mixed Model forest) using (a) all 46 predictors, (b) a variable selection algorithm, and (c) the top 5 most important predictors. Age was included in all models. Performance was evaluated using area under the receiver operating curve in 2 internal validation data sets. RESULTS: Area under the receiver operating curve ranged from 0.80 to 0.84 for the models. The most important predictors of mobility limitation were ease of getting up from a chair, gait speed, self-reported health status, body mass index, and depression. CONCLUSIONS: Machine learning models using repeated measures had good performance for identifying older adults at risk of developing mobility limitation. Future studies should evaluate the utility and efficiency of the prediction models as a tool in clinical settings for identifying at-risk older adults who may benefit from interventions aimed to prevent or delay mobility limitation.

Detection of Cognitive Impairment after Critical Illness with the Medicare Annual Wellness Visit: A Cohort Study.

Rationale: Cognitive impairment after critical illness is common in observational studies of older intensive care unit (ICU) survivors. The rate of screening for and diagnosis of cognitive impairment in ICU survivors in nonresearch settings is unknown. Objectives: To determine how often cognitive impairment was detected in older adults in the year after critical illness at an academic medical center as part of 1) the Medicare Annual Wellness Visit (AWV) and 2) routine clinical care. Methods: This study was a retrospective cohort study conducted at an urban academic medical center. The study included 696 patients aged 65 years and older admitted to the medical ICU between October 1, 2016, and October 1, 2018, and discharged alive. Patients were also required to have a health system-affiliated primary care provider. Patients were followed for 1 year. We defined cognitive impairment detected in the AWV as either an indicated diagnosis of cognitive impairment or dementia or patient, family, or provider indication of memory concerns during the AWV. We modeled the incidence of AWV completion and the detection of cognitive impairment using semiparametric additive models accounting for the competing risk of death. Results: Over 1 year of follow-up, the cumulative incidence of mortality was 23.0% (95% confidence interval [CI], 19.9-26.1%), with 24.7% (95% CI, 21.5-27.9%) completing the AWV. The cumulative incidence of cognitive impairment first detected through the AWV was 3.4% (95% CI, 1.8-5.0%) at 1 year, with a higher cumulative incidence for diagnoses of cognitive impairment or dementia first indicated via encounter diagnosis codes or the electronic health record problem list (5.9%; 95% CI, 3.9-7.9%). Conclusions: The results of our study suggest that the currently implemented AWV is unlikely to be an adequate mechanism for detecting cognitive impairment in a high-risk population such as those recovering from critical illness.

Effectiveness of a Nurse-Led Multidisciplinary Intervention vs Usual Care on Advance Care Planning for Vulnerable Older Adults in an Accountable Care Organization: A Randomized Clinical Trial.

Importance: Advance care planning (ACP), especially among vulnerable older adults, remains underused in primary care. Additionally, many ACP initiatives fail to integrate directly into the electronic health record (EHR), resulting in infrequent and disorganized documentation. Objective: To determine whether a nurse navigator-led ACP pathway combined with a health care professional-facing EHR interface improves the occurrence of ACP discussions and their documentation within the EHR. Design, Setting, and Participants: This was a randomized effectiveness trial using the Zelen design, in which patients are randomized prior to informed consent, with only those randomized to the intervention subsequently approached to provide informed consent. Randomization began November 1, 2018, and follow-up concluded November 1, 2019. The study population included patients 65 years or older with multimorbidity combined with either cognitive or physical impairments, and/or frailty, assessed from 8 primary care practices in North Carolina. Interventions: Participants were randomized to either a nurse navigator-led ACP pathway (n = 379) or usual care (n = 380). Main Outcomes and Measures: The primary outcome was documentation of a new ACP discussion within the EHR. Secondary outcomes included the usage of ACP billing codes, designation of a surrogate decision maker, and ACP legal form documentation. Exploratory outcomes included incident health care use. Results: Among 759 randomized patients (mean age 77.7 years, 455 women [59.9%]), the nurse navigator-led ACP pathway resulted in a higher rate of ACP documentation (42.2% vs 3.7%, P < .001) as compared with usual care. The ACP billing codes were used more frequently for patients randomized to the nurse navigator-led ACP pathway (25.3% vs 1.3%, P < .001). Patients randomized to the nurse navigator-led ACP pathway more frequently designated a surrogate decision maker (64% vs 35%, P < .001) and completed ACP legal forms (24.3% vs 10.0%, P < .001). During follow-up, the incidence of emergency department visits and inpatient hospitalizations was similar between the randomized groups (hazard ratio, 1.17; 95% CI, 0.92-1.50). Conclusions and Relevance: A nurse navigator-led ACP pathway integrated with a health care professional-facing EHR interface increased the frequency of ACP discussions and their documentation. Additional research will be required to evaluate whether increased EHR documentation leads to improvements in goal-concordant care. Trial Registration: ClinicalTrials.gov Identifier: NCT03609658.

Automated Frailty Screening At-Scale for Pre-Operative Risk Stratification Using the Electronic Frailty Index.

BACKGROUND: Frailty is associated with numerous post-operative adverse outcomes in older adults. Current pre-operative frailty screening tools require additional data collection or objective assessments, adding expense and limiting large-scale implementation. OBJECTIVE: To evaluate the association of an automated measure of frailty integrated within the Electronic Health Record (EHR) with post-operative outcomes for nonemergency surgeries. DESIGN: Retrospective cohort study. SETTING: Academic Medical Center. PARTICIPANTS: Patients 65 years or older that underwent nonemergency surgery with an inpatient stay 24 hours or more between October 8th, 2017 and June 1st, 2019. EXPOSURES: Frailty as measured by a 54-item electronic frailty index (eFI). OUTCOMES AND MEASUREMENTS: Inpatient length of stay, requirements for post-acute care, 30-day readmission, and 6-month all-cause mortality. RESULTS: Of 4,831 unique patients (2,281 females (47.3%); mean (SD) age, 73.2 (5.9) years), 4,143 (85.7%) had sufficient EHR data to calculate the eFI, with 15.1% categorized as frail (eFI > 0.21) and 50.9% pre-frail (0.10 < eFI ≤ 0.21). For all outcomes, there was a generally a gradation of risk with higher eFI scores. For example, adjusting for age, sex, race/ethnicity, and American Society of Anesthesiologists class, and accounting for variability by service line, patients identified as frail based on the eFI, compared to fit patients, had greater needs for post-acute care (odds ratio (OR) = 1.68; 95% confidence interval (CI) = 1.36-2.08), higher rates of 30-day readmission (hazard ratio (HR) = 2.46; 95%CI = 1.72-3.52) and higher all-cause mortality (HR = 2.86; 95%CI = 1.84-4.44) over 6 months' follow-up. CONCLUSIONS: The eFI, an automated digital marker for frailty integrated within the EHR, can facilitate pre-operative frailty screening at scale.

Embedding and Sustaining a Focus on Function in Specialty Research and Care.

Function and the independent performance of daily activities are of critical importance to older adults. Although function was once a domain of interest primarily limited to geriatricians, transdisciplinary research has demonstrated its value across the spectrum of medical and surgical care. Nonetheless, integrating a functional perspective into medical and surgical therapeutics has yet to be implemented consistently into clinical practice. This article summarizes the presentations and discussions from a workshop, "Embedding/Sustaining a Focus on Function in Specialty Research and Care," held on January 31 to February 1, 2019. The third in a series supported by the National Institute on Aging and the John A. Hartford Foundation, the workshop aimed to identify scientific gaps and recommend research strategies to advance the implementation of function in care of older adults. Transdisciplinary leaders discussed implementation of mobility programs and functional assessments, including comprehensive geriatric assessment; integrating cognitive and sensory functional assessments; the role of culture, environment, and community in incorporating function into research; innovative methods to better identify functional limitations, techniques, and interventions to facilitate functional gains; and the role of the health system in fostering integration of function. Workshop participants emphasized the importance of aligning goals and assessments and adopting a team science approach that includes clinicians and frontline staff in the planning, development, testing, and implementation of tools and initiatives. This article summarizes those discussions.

Machine Learning in Aging: An Example of Developing Prediction Models for Serious Fall Injury in Older Adults.

BACKGROUND: Advances in computational algorithms and the availability of large datasets with clinically relevant characteristics provide an opportunity to develop machine learning prediction models to aid in diagnosis, prognosis, and treatment of older adults. Some studies have employed machine learning methods for prediction modeling, but skepticism of these methods remains due to lack of reproducibility and difficulty in understanding the complex algorithms that underlie models. We aim to provide an overview of two common machine learning methods: decision tree and random forest. We focus on these methods because they provide a high degree of interpretability. METHOD: We discuss the underlying algorithms of decision tree and random forest methods and present a tutorial for developing prediction models for serious fall injury using data from the Lifestyle Interventions and Independence for Elders (LIFE) study. RESULTS: Decision tree is a machine learning method that produces a model resembling a flow chart. Random forest consists of a collection of many decision trees whose results are aggregated. In the tutorial example, we discuss evaluation metrics and interpretation for these models. Illustrated using data from the LIFE study, prediction models for serious fall injury were moderate at best (area under the receiver operating curve of 0.54 for decision tree and 0.66 for random forest). CONCLUSIONS: Machine learning methods offer an alternative to traditional approaches for modeling outcomes in aging, but their use should be justified and output should be carefully described. Models should be assessed by clinical experts to ensure compatibility with clinical practice.