Wilson, Patrick M.

Context: Early palliative care can improve end-of-life outcomes, but referrals to palliative care specialists can be delayed in the primary care setting. Objective: This pilot study assessed the effect of a machine-learning algorithm on time to palliative care in a primary care population. Methods: Patients (aged ≥18 years) were eligible if they were empaneled with a primary care provider (PCP) from July 20, 2020, through May 30, 2021. The algorithm evaluated their health records and presented patients who were predicted to have the greatest need for palliative care. Records were then reviewed by palliative care specialists, and patients were randomized in a stepped-wedge fashion to have a referral notification sent to their PCPs if unmet palliative care needs were verified. Time-to-event outcomes were evaluated with Poisson regression models. Results: Of the 127,080 patients evaluated, 934 had their health records presented for review. Some patients were repeatedly presented by the algorithm (total presentations: 1592). In the intervention arm, PCPs were prompted to order a palliative care consultation for 142 patients. The time to 0.1% of the population receiving a palliative care consultation was 60.9 days for the intervention arm versus 71.8 days for the control arm (probability of a shorter time with the intervention, 0.88). Conclusion: A machine-learning algorithm to identify palliative care needs was successfully integrated into a primary care practice. More work is needed to improve the workflow.

1559 Background: Timely access to specialty Palliative Care (SPC) services provides significant benefits to hospitalized patients with cancer, including improvements in quality of life. However, many patients with cancer who could benefit from SPC suffer from lack of timely referral or do not receive SPC services. A previously published trial of an artificial intelligence/machine learning (AI/ML) model to predict need for SPC at Mayo Clinic increased timely SPC consultation of hospitalized patients overall. We sought to review the performance of this algorithm in patients with cancer across an expanded hospitalized population. Methods: The study population consisted of all patients admitted into a Mayo Clinic hospital in Minnesota, Wisconsin, Arizona and Florida between January 2020 and September 2023. Due to the size of the cohort a case-control sample of three controls for every case was created. Patients were considered to be a patient with cancer if they had at least one billing diagnosis up to 1 year prior to their index hospitalization from any of the 5 HCC categories denoting cancer: Metastatic Cancer and Acute Leukemia (HCC 8); Lung and Other Severe Cancers (HCC 9); Lymphoma and Other Cancers (HCC 10); Colorectal, Bladder, and Other Cancers (HCC 11); and Breast, Prostate, and Other Cancers and Tumors (HCC 12). The training data set consisted of 107,076 patient encounters with a total of 8,355,090 time periods of constant risk. An AI/ML model using gradient boosting methods which contained 269 variables (both static and time-varying) of various classes with SPC consultation treated as a time-to-event outcome was trained. Results: Due to the longitudinal nature of the prediction, performance was assessed using the max score AUC (using the max score a patient received during a given encounter prior to an event or discharge) to produce the AUC. The model had an overall AUC of 0.932 (0.929, 0.935 – 95% CI). We saw a decrease in performance of the AI/ML model in the Oncology population with an overall performance of 0.818 (0.806, 0.83). Some of the most influential variables were previous SPC visit, age, metastatic disease, acute leukemia diagnosis, and pain scores. Conclusions: An AI/ML model can effectively predict the need for an inpatient PC consult in a population of hospitalized patients with cancer. [Table: see text]

AbstractIntroductionWe investigated the association of the area deprivation index (ADI) with cognitive decline, mild cognitive impairment (MCI), and dementia in older adults (≥50 years old). ADI is a neighborhood socioeconomic disadvantage measure assessed at the census block group level.MethodsThe study included 4699 participants, initially without dementia, with available ADI values for 2015 and at least one study visit in 2008 through 2018. Using logistic regression and Cox proportional hazards models with age as the time scale, we assessed the odds for MCI and the risk for dementia, respectively.ResultsIn cognitively unimpaired (CU) adults at baseline, the risk for progression to dementia increased for every decile increase in the ADI state ranking (hazard ratio = 1.06, 95% confidence interval (1.01–1.11), P = .01). Higher ADI values were associated with subtly faster cognitive decline.DiscussionIn older CU adults, higher baseline neighborhood socioeconomic deprivation levels were associated with progression to dementia and slightly faster cognitive decline.Highlights The study used area deprivation index, a composite freely available neighborhood deprivation measure. Higher levels of neighborhood deprivation were associated with greater mild cognitive impairment odds. Higher neighborhood deprivation levels were associated with higher dementia risk.

Abstract Background: Palliative care is a medical specialty centered on improving the quality of life (QOL) of patients with complex or life-threatening illnesses. The need for Palliative care is increasing and with that the rigorous testing of triage tools that can be used quickly and reliably to identify patients that may benefit from palliative care. Methods: To that Aim we will conduct a two-armed stepped-wedge cluster randomized trial rolled out to two inpatient hospitals to evaluate whether a machine learning algorithm accurately identifies patients who may benefit from a comprehensive review by a palliative care specialist, and decreases time to receiving a palliative care consult in hospital. This is a single-center study which will be conducted from August 2019 to November 2020 at Saint Mary’s Hospital & Methodist Hospital, Mayo Clinic, Rochester, Minnesota. Clusters will be nursing units which will be chosen to be a mix of complex patients from Cardiology, Critical Care and Oncology and had previously established relationships with palliative medicine. The stepped wedge design will have 12 units allocated to a design matrix of 5 treatment wedges. Each wedge will last 75 days resulting in a study period of 12 months unless otherwise specified. Data will be analyzed with Bayesian hierarchical models with credible intervals denoting statistical significance.Discussion: This intervention offers a pragmatic approach to delivering specialty palliative care to hospital patients in need using machine learning, thereby leading to high value care and improved outcomes. It is not enough for AI to be utilized by simply publishing research showing predictive performance; Clinical trials demonstrating better outcomes are critically needed. Furthermore, the deployment of an AI algorithm is a complex process that requires multiple teams with varying skill sets. To evaluate a deployed AI, a pragmatic clinical trial can accommodate the difficulties of clinical practice whilst retaining scientific rigor.Trial registration NCT03976297 (ClinicalTrials.gov), registered on June 6, 2019 prior to trial start.

Abstract Objective Access to palliative care (PC) is important for many patients with uncontrolled symptom burden from serious or complex illness. However, many patients who could benefit from PC do not receive it early enough or at all. We sought to address this problem by building a predictive model into a comprehensive clinical framework with the aims to (i) identify in-hospital patients likely to benefit from a PC consult, and (ii) intervene on such patients by contacting their care team. Materials and Methods Electronic health record data for 68 349 inpatient encounters in 2017 at a large hospital were used to train a model to predict the need for PC consult. This model was published as a web service, connected to institutional data pipelines, and consumed by a downstream display application monitored by the PC team. For those patients that the PC team deems appropriate, a team member then contacts the patient’s corresponding care team. Results Training performance AUC based on a 20% holdout validation set was 0.90. The most influential variables were previous palliative care, hospital unit, Albumin, Troponin, and metastatic cancer. The model has been successfully integrated into the clinical workflow making real-time predictions on hundreds of patients per day. The model had an “in-production” AUC of 0.91. A clinical trial is currently underway to assess the effect on clinical outcomes. Conclusions A machine learning model can effectively predict the need for an inpatient PC consult and has been successfully integrated into practice to refer new patients to PC.

Objective: To understand the interaction of multimorbidity and functional limitations in determining health-care utilization and survival in older adults. Methods: Olmsted County, Minnesota, residents aged 60–89 years in 2005 were categorized into four cohorts based on the presence or absence of multimorbidity (≥3 chronic conditions from a list of 18) and functional limitations (≥1 limitation in an activity of daily living from a list of 9), and were followed through December 31, 2016. Andersen–Gill and Cox regression estimated hazard ratios (HRs) for emergency department (ED) visits, hospitalizations, and death using persons with neither multimorbidity nor functional limitations as the reference (interaction analyses). Results: Among 13,145 persons, 34% had neither multimorbidity nor functional limitations, 44% had multimorbidity only, 4% had functional limitations only, and 18% had both. Over a median follow-up of 11 years, 5906 ED visits, 2654 hospitalizations, and 4559 deaths occurred. Synergistic interactions on an additive scale of multimorbidity and functional limitations were observed for all outcomes; however, the magnitude of the interactions decreased with advancing age. The HR (95% confidence interval) for death among persons with both multimorbidity and functional limitations was 5.34 (4.40–6.47) at age 60–69, 4.16 (3.59–4.83) at age 70–79, and 2.86 (2.45–3.35) at age 80–89 years. Conclusion: The risk of ED visits, hospitalizations, and death among persons with both multimorbidity and functional limitations is greater than additive. The magnitude of the interaction was strongest for the youngest age group, highlighting the importance of interventions to prevent and effectively manage multimorbidity and functional limitations early in life.

PURPOSE: We aimed to better understand how similarly patients with colorectal cancer and caregivers view care quality and to assess factors that may influence concordance. MATERIALS AND METHODS: We conducted a secondary analysis of paired patient and caregiver quality ratings of colorectal cancer care in three specific domains: surgery, chemotherapy overall, and chemotherapy nursing. Agreement was assessed with difference scores, concordance with Gwet second-order agreement statistics (AC2), and variation in agreement with stratified analyses. We examined whether the care experiences of patients and caregivers were associated with top-box (most-positive) ratings and examined variations in concordance on the basis of the presence of a top-box score. RESULTS: Four hundred seventeen patient-caregiver dyads completed the surveys. Quality-of-care ratings were positively skewed, with most dyads indicating top-box ratings. Patient and caregiver care experiences were highly associated with top-box ratings. Overall patient-caregiver concordance was very high for all three care domains (surgery: AC2, 0.87 [95% CI, 0.83 to 0.90]; chemotherapy overall: AC2, 0.84 [95% CI, 0.79 to 0.88]; chemotherapy nursing: AC2, 0.91 [95% CI, 0.87 to 0.94]). Stratified analyses of patient and caregiver characteristics did not identify any patterns that consistently affected concordance. The concordance statistic significantly decreased for all three outcomes ( P < .001), however, when the patient or caregiver assessed quality as anything other than top box. CONCLUSION: Caregiver and patient reports on care quality were highly concordant for top-box care and did not vary with patient or caregiver factors. Additional exploration is needed to identify reasons for increased variability when the quality scores were less than a top-box response.

AbstractBackgroundReflecting (“stop‐and‐think”) before rating may help patients consider the quality of shared decision making (SDM) and mitigate ceiling/halo effects that limit the performance of self‐reported SDM measures.MethodsWe asked a diverse patient sample from the United States to reflect on their care before completing the 3‐item CollaboRATE SDM measure. Study 1 focused on rephrasing CollaboRATE items to promote reflection before each item. Study 2 used 5 open‐ended questions (about what went well and what could be improved upon, signs that the clinician understood the patient's situation, how the situation will be addressed, and why this treatment plan makes sense) to invite reflection before using the whole scale. A linear analogue scale assessed the extent to which the plan of care made sense to the patient.ResultsIn Study 1, 107 participants completed surveys (84% response rate), 43 (40%) rated a clinical decision of which 27 (63%) after responding to reflection questions. Adding reflection lowered CollaboRATE scores (“less” SDM) and reduced the proportion of patients giving maximum (ceiling) scores (not statistically significant). In Study 2, 103 of 212 responders (49%) fully completed the version containing reflection questions. Reflection did not significantly change the distribution of CollaboRATE scores or of top scores. Participants indicated high scores on the sense of their care plan (mean 9.7 out of 10, SD 0.79). This rating was weakly correlated with total CollaboRATE scores (rho = .4, P = .0001).ConclusionReflection‐before‐quantification interventions may not improve the performance of patient‐reported measures of SDM with substantial ceiling/halo effects.

Background: In 2012, updated cervical cancer screening recommendations were released with consensus on Papanicolaou (Pap) testing every 3 years for women age 21–65 years or Pap–human papillomavirus (HPV) cotesting at 5-year intervals for women age 30–65 years. Primary study aims: Assess current use of Pap-HPV cotesting and describe local population trends over time in Pap and Pap-HPV cotesting. Secondary aim: Assess sociodemographic factors correlating with screening. Methods: We assessed Rochester Epidemiology Project data for Pap and Pap-HPV cotesting among women age 16 years and older living in Olmsted County, Minnesota, yearly from 2005 (study population n = 47,203) through 2016 (study population n = 49,510). We calculated 3-year (Pap) and 5-year (Pap-HPV) moving prevalence rates of screening as proportion of eligible population. Multivariable logistic regression was used to assess factors potentially associated with screening. Results: In 2016, 64.6% of 27,418 eligible 30- to 65-year-old women were up to date with cervical cancer screening; 60.8% had received Pap-HPV cotest screening. Significant declines in Pap completion rates over time were observed in all age groups, including an unexpected decline in 21- to 29-year-old women. Coincident with decreasing Pap screening rates, Pap-HPV cotesting significantly increased among women age 30–65 years, from 10.0% in 2007 to 60.8% in 2016. Conclusions: This suggests increasing adoption of 2012 screening recommendations in the 30- to 65-year-old population. However, decline in Pap screening among 21- to 29-year-old women is concerning. Disparities by race, ethnicity, smoking status, and comorbidity level were observed. Results suggest need for multilevel patient and clinician interventions to increase cervical cancer screening adherence.