Adjusted Clinical Groups Research Articles

Validity of the Johns Hopkins Adjusted Clinical Groups system on the utilisation of healthcare services in Norway: a retrospective cross-sectional study

BackgroundThe Adjusted Clinical Groups (ACG) System is a validated electronic risk stratification system. However, there is a lack of studies on the association between different ACG risk scores and the utilisation of different healthcare services using different sources of input data. The aim of this study was therefore to assess the validity of the association between five different ACG risk scores and the utilisation of a range of different healthcare services using input data from either general practitioners (GPs) or hospitals.MethodsRegistry-based study of all adult inhabitants in four Norwegian municipalities that received somatic healthcare in one year (N = 168 285). The ACG risk scores resource utilisation band, unscaled ACG concurrent risk, unscaled concurrent risk, frailty flag and chronic condition count were calculated using age, sex and diagnosis codes from GPs and a hospital, respectively. Healthcare utilisation covered GP, municipal and hospital services. Areas under the receiver operating curve (AUC) were calculated and compared to the AUC of a model using only age and sex.ResultsUtilisation of all healthcare services increased with increasing scores in the “resource utilisation band” (RUB) and all other investigated ACG risk scores. The risk scores overall distinguished well between levels of utilisation of GP visits (AUC up to 0.84), hospitalisation (AUC up to 0.8) and specialist outpatient visits (AUC up to 0.72), but not out-of-hours GP visits (AUC up to 0.62). The score “unscaled ACG concurrent risk” overall performed best. Risk scores based on data from either GPs or hospitals performed better for the classification of healthcare services in their respective domains. The model based on age and sex performed better for distinguishing between levels of utilisation of municipal services (AUC 0.83–0.90 compared to 0.46–0.79).ConclusionsRisk scores from the ACG system is valid for classifying GP visits, hospitalisation and specialist outpatient visits. It does not outperform simpler models in the classification of utilisation of municipal services such as nursing homes and home services and outpatient emergency care in primary healthcare. The ACG system can be applied in Norway using administrative data from either GPs or hospitals.

Lichen Sclerosus is a high-cost diagnosis, compounded by comorbid conditions

Introduction and Objective: In the past decade, there have been significant advancements in understanding the etiology, molecular mechanisms, and best treatment options for lichen sclerosus (LS). However, to date, there is a lack of published data on the costs of care, associated comorbidities, and quality of life related to the condition. Therefore, this study aims to investigate the global costs associated with the treatment of LS using a contained insurance dataset. Methods: Deidentified 12-month data from 2022 to 2023 for a northeastern private insurer of teachers and municipal workers was queried to analyze costs associated with a diagnosis of LS (ICD-10 code L90.0). Comorbidities were also queried. Descriptive statistics were utilized to describe the population’s expenditures. Johns Hopkins Adjusted Clinical Groups (ACG) risk assessment was utilized to determine future costs. Z-tests were utilized to determine the differences in rates of comorbidities, and T-tests for differences in cost. Results: The database included all 4,306 members. Average annual cost per member was $9,280, average age was 35.5 years, behavioral health prevalence was 26%, and 53% were female. Cost for members with a behavioral health condition was $18,125 (2.0-fold) and $6,219 average cost for members without (0.7-fold). LS was present in 161 (3.7%) members, with average age of 42.0, prevalence of behavioral health 37%, and 62% were female. The presence of lichen sclerosus was associated with a 3.5-fold increase in average cost per member ($32,437 annually). For those with LS and a behavioral health condition, cost was $53,539 (5.8-fold average per member) compared to $19,901 (2.1-fold average plan cost) for those with LS and no BH condition. Median cost for those with LS was $8,755 compared to $2,056 for the entire cohort which was a significant difference (P < 0.0001). Incidence of behavioral health conditions, low back pain, obesity, chronic obstructive pulmonary disease, hyperlipidemia and hypertension were increased in the LS group versus the plan population (P < 0.01). Pharmaceutical expenditure for those with LS was a median of $267.58 compared to $94.80 average for the entire cohort (P < 0.0001). Conclusions: The presence of LS in this privately insured population significantly raises costs, exacerbated by elevated rates of mental health and other comorbidities, highlighting the need for comprehensive care addressing these complexities to improve health outcomes and reduce overall expenses.

Utility of combining frailty and comorbid disease indices in predicting outcomes following craniotomy for adult primary brain tumors: A mixed-effects model analysis using the nationwide readmissions database

ObjectiveThe escalating healthcare expenditures in the United States, particularly in neurosurgery, necessitate effective tools for predicting patient outcomes and optimizing resource allocation. This study explores the utility of combining frailty and comorbidity indices, specifically the Johns Hopkins Adjusted Clinical Groups (JHACG) frailty index and the Elixhauser Comorbidity Index (ECI), in predicting hospital length of stay (LOS), non-routine discharge, and one-year readmission in patients undergoing craniotomy for benign and malignant primary brain tumors. MethodsLeveraging the Nationwide Readmissions Database (NRD) for 2016–2019, we analyzed data from 645 patients with benign and 30,991 with malignant tumors. Frailty, ECI, and frailty + ECI were assessed as predictors using generalized linear mixed-effects models. Receiver operating characteristic (ROC) curves evaluated predictive performance. ResultsPatients in the benign tumor cohort had a mean LOS of 8.1 ± 15.1 days, and frailty + ECI outperformed frailty alone in predicting non-routine discharge (AUC 0.829 vs. 0.820, p = 0.035). The malignant tumor cohort patients had a mean LOS of 7.9 ± 9.1 days. In this cohort, frailty + ECI (AUC 0.821) outperformed both frailty (AUC 0.744, p < 0.0001) and ECI alone (AUC 0.809, p < 0.0001) in predicting hospital LOS. Frailty + ECI (AUC 0.831) also proved superior to frailty (AUC 0.809, p < 0.0001) and ECI alone (AUC 0.827, p < 0.0001) in predicting non-routine discharge location for patients with malignant tumors. All indices performed comparably to one another as a predictor of readmission in both cohorts. ConclusionThis study highlights the synergistic predictive capacity of frailty + ECI, especially in malignant tumor cases, and further suggests that comorbid diseases may greatly influence perioperative outcomes more than frailty. Enhanced risk assessment could aid clinical decision-making, patient counseling, and resource allocation, ultimately optimizing patient outcomes.

Towards Regional Population Health Management: A Prospective Analysis Using the Adjusted Clinical Groups Classification.

This research seeks to assess the potential of regionally integrated health management for specific sub-populations, including the incorporation of self-management initiatives. It will achieve this by conducting a thorough stratification analysis of hospital data, utilizing the Adjusted Clinical Groups (ACG) classification system. The approach involves a retrospective review of healthcare data spanning five years, which includes patient demographics, health outcomes, and healthcare utilization metrics. We intend to use the ACG method to classify the patient population into pertinent groups that mirror their health requirements and resource use. The insights obtained from this analysis will be used to create a localized adaptation of the Kaiser Permanente Pyramid Model of Care. This adaptation aims to identify the distribution of costs among patients treated in the Rivierenland Hospital. We anticipate that stratifying data with the ACG method will identify distinct multimorbid subgroups. These subgroups will have unique healthcare requirements. Early interventions and customized health management strategies, based on these insights, could enhance health outcomes and resource efficiency for high-risk patients. This analysis will serve as a foundation for constructive discussions with hospital management and clinical staff, fostering a deeper comprehension of the patients' burden of disease. It might also foster multidisciplinary collaboration opportunities between medical specialties as with regional healthcare partners such as general practitioners (GPs), mental health and other long-term care organizations. Moreover, we anticipate that self-care initiatives, supported by customized health information, will encourage increased patient engagement and strategies for enhancing lifestyle improvements. This strategy is expected to enable the personalization of advanced care planning based on individual needs profiles, thereby improving the management of complex and chronic conditions, and encouraging self-care practices. Our anticipated findings highlight the potential benefits of a data-informed approach to advancing healthcare outcomes and present opportunities for future investigations to refine and implement such integrated care models across the region.

Predicting Provider Workload Using Predicted Patient Risk Score and Social Determinants of Health in Primary Care Setting.

Provider burnout due to workload is a significant concern in primary care settings. Workload for primary care providers encompasses both scheduled visit care and non-visit care interactions. These interactions are highly influenced by patients' health conditions or acuity, which can be measured by the Adjusted Clinical Group (ACG) score. However, new patients typically have minimal health information beyond social determinants of health (SDOH) to determine ACG score. This study aims to assess new patient workload by first predicting the ACG score using SDOH, age, and gender and then using this information to estimate the number of appointments (scheduled visit care) and non-visit care interactions. Two years of appointment data were collected for patients who had initial appointment requests in the first year and had the ACG score, appointment, and non-visit care counts in the subsequent year. State-of-art machine learning algorithms were employed to predict ACG scores and compared with current baseline estimation. Linear regression models were then used to predict appointments and non-visit care interactions, integrating demographic data, SDOH, and predicted ACG scores. The machine learning methods showed promising results in predicting ACG scores. Besides the decision tree, all other methods performed at least 9% better in accuracy than the baseline approach which had an accuracy of 78%. Incorporating SDOH and predicted ACG scores also significantly improved the prediction for both appointments and non-visit care interactions. The R 2 values increased by 95.2 and 93.8%, respectively. Furthermore, age, smoking tobacco, family history, gender, usage of injection birth control, and ACG were significant factors for determining appointments. SDOH factors such as tobacco usage, physical exercise, education level, and group activities were strongly correlated with non-visit care interactions. The study highlights the importance of SDOH and predicted ACG scores in predicting provider workload in primary care settings.

Integration of Chronological Age Does Not Improve the Performance of a Mixed-Effect Model Using Comorbidity Burden and Frailty to Predict 90-Day Readmission After Surgery for Degenerative Scoliosis

ObjectiveWe evaluated the contributions of chronological age, comorbidity burden, and/or frailty in predicting 90-day readmission in patients undergoing degenerative scoliosis surgery. MethodsPatients were identified through the Healthcare Cost and Utilization Project Nationwide Readmissions Database. Frailty was assessed using the Johns Hopkins Adjusted Clinical Groups frailty-defining indicator. Comorbidity was assessed using the Elixhauser Comorbidity Index (ECI). Generalized linear mixed-effects models were created to predict readmission using age, frailty, and/or ECI. Area under the curve (AUC) were compared using DeLong's test. Results8,104 patients were identified. Readmission rate was 9.8%, with infection representing the most common cause (3.5%). Our first model utilized chronological age, ECI, and/or frailty as primary predictors. The combination of ECI+frailty+age performed best, but the inclusion of chronological age did not significantly improve performance compared to ECI+frailty alone (AUC 0.603 vs. 0.599, p=0.290). A second model using only chronological age and frailty as primary predictors performed better, however the inclusion of chronological age worsened performance when compared to frailty alone (AUC 0.747 vs. 0.743, p=0.043). ConclusionsThese data support frailty as a predictor of 90-day readmission within a nationally representative sample. Frailty alone performed better than combinations of ECI and age. Interestingly, the integration of chronological age did not dramatically improve the model’s performance. Limitations include the use of a national registry and a single frailty index. This provides impetus to explore biological age, rather than chronological age, as a potential tool for surgical risk assessment.

174 Developing Mixed-Effects Models to Optimize Prediction of Postoperative Outcomes in a Modern Sample of Over 450,000 Patients Undergoing Elective Cervical Spine Fusion Surgery

INTRODUCTION: Cervical spine decompression and fusion surgery is commonly used to treat degenerative cervical spine pathologies.However, there is a paucity of high-quality data defining the optimal comorbidity indices specifically in patients undergoing cervical spine fusion surgery. METHODS: Using data from 2016–2019, we queried the Nationwide Readmissions Database (NRD) to identify individuals who had received cervical spine fusion surgery. The Johns Hopkins Adjusted Clinical Groups (JHACG) frailty-defining indicator was used to assess patient frailty. To measure the level of comorbidity, Elixhauser Comorbidity Index (ECI) scores were queried. ROC curves were developed utilizing comorbidity indices as predictor variables for pertinent complications such as death, non-routine discharge, top-quartile cost, top-quartile length of stay (LOS), and one-year readmission. RESULTS: A total of 453,717 patients were eligible for analysis. Non-routine discharges occurred in 93,961 (20.7%) patients. The mean adjusted all-payer cost for the procedure was $22,573.14 ± $18,274.86 (Top Quartile: $26,775.80) and the mean LOS was 2.7 days ± 4.4 days (Top Quartile: 4.7 days). After surgery, there were 703 (0.15%) mortalities and 58,254 (12.8%) readmissions within one year post-operatively. Models using Frailty+ECI as the primary predictor consistently outperformed the ECI only model with statistically significant p-values, as determined by comparing respective AUCs, for most of the complications assessed. Cost and mortality were the only outcomes for which this was not the case, as Frailty outperformed both ECI and Frailty+ECI in cost (p &lt; 0.0001 for all) and Frailty+ECI preformed as well as ECI alone in mortality (p = 0.10). CONCLUSIONS: Our data suggests that Frailty+ECI may most accurately predict clinical outcomes in patients receiving cervical spine surgery. These models may be used to identify high risk populations and patients who may necessitate greater resource utilization.

Off-Pump Coronary Artery Bypass Grafting Does Not Confer Superior Outcomes Among Frail Patients

Off-pump coronary revascularization (OPCAB) has been proposed to benefit patients who are at a greater surgical risk because it avoids the use of extracorporeal circulation. Although, historically, older patients were considered high-risk candidates, recent studies implicate frailty as a more comprehensive measure of perioperative fitness. Yet, the outcomes of OPCAB in frail patients have not been elucidated. Thus, using a national cohort of frail patients, we assessed the impact of OPCAB relative to on-pump coronary revascularization (ONCAB). Patients who underwent first-time elective coronary revascularization were tabulated from the 2010 to 2020 Nationwide Readmissions Database. Frailty was assessed using the previously-validated Johns Hopkins Adjusted Clinical Groups indicator. Multivariable models were used to consider the independent associations between OPCAB and the key outcomes. Of ∼26,529 frail patients, 6,322 (23.8%) underwent OPCAB. After risk adjustment and compared with ONCAB, OPCAB was linked with similar odds of in-hospital mortality but greater likelihood of postoperative cardiac arrest (adjusted odds ratio [AOR] 1.53, confidence interval [CI] 1.13 to 2.07) and myocardial infarction (AOR 1.44, CI 1.23 to 1.69). OPCAB was further associated with greater odds of postoperative infection (AOR 1.22, CI 1.02 to 1.47) but decreased need for blood transfusion (AOR 0.68, CI 0.60 to 0.77). In addition, OPCAB faced a +0.86-day increase in length of stay (CI 0.21 to 1.51) but similar costs (β $1,610, CI -$1,240 to 4,460) relative to ONCAB. Although OPCAB was associated with no difference in mortality compared with ONCAB, it was linked with greater likelihood of postoperative cardiac arrest and myocardial infarction. Our findings demonstrate that ONCAB remains associated with superior outcomes, even in the growing population of frail patients who underwent coronary revascularization.

OP173 Estimating The Marginal Productivity Of Health Technology Adoption

IntroductionDecisions to adopt health technologies rely, in part, on judgements about cost effectiveness. Cost effectiveness is commonly assessed against a willingness-to-pay threshold for health gains. Building an evidence base on the marginal productivity of health spending to inform the value of the threshold is increasingly of interest for resource allocation decision-making and technology implementation. We report on an in-progress analysis to inform a threshold for policy purposes in British Columbia, Canada.MethodsWe developed a ten-year panel-data model with instrumental variables, which lessens the degree of time-invariant confounding and addresses biased causal inferences caused by unobserved factors, to provide estimates of the marginal cost per health unit measured using quality-adjusted life-years (QALYs). We use the Johns Hopkins Adjusted Clinical Group (ACG) system and a British Columbia Health System Matrix to classify patients into six resource use bands (RUBs) ranging from ‘healthy’ to ‘very high morbidity’. Patients are also classified by chronic conditions and types of services. Place of residence and geographical region of health authorities are considered. Variables included age, gender, mortality and comorbidity rates, costs of hospitalizations, emergency department and physician visits, residential and home care, laboratory services, diagnosis and medications, and quality of life. Instrumental variables included sociodemographic characteristics as reported in the Canadian census.ResultsThe largest RUB was ‘moderate’ morbidity (39.3%), while the smallest was ‘healthy’ (1.5%). The youngest was the ‘low’ morbidity (mean 31, standard deviation [SD] 21) and the oldest was ‘very high’ (mean 69, SD 17). The healthy group had the smallest mean costs (CND563, SD CND4,121; equivalent to USD421, SD USD3,083). In contrast, the ‘very high’ group had the largest (CND20,398, SD CND36,188; equivalent to USD15,258, SD USD27,069). Age and gender standardized comorbidity index scores ranged from 0.05 to 6.41 (median 0.98). Additional analyses (e.g., costs per QALY) are ongoing and the results will be reported at the conference.ConclusionsOur empirical approach is robust and flexible, allowing estimates of marginal productivity according to factors such as disease, geographical region, service type, and care sector. This work has applications at the provincial and national levels and adds to methodological literature in the field.

Validating and Improving Adjusted Clinical Group's Future Hospitalization and High-Cost Prediction Models for Dutch Primary Care.

The rise in health care costs, caused by older and more complex patient populations, requires Population Health Management approaches including risk stratification. With risk stratification, patients are assigned individual risk scores based on medical records. These patient stratifications focus on future high costs and expensive care utilization such as hospitalization, for which different models exist. With this study, the research team validated the accuracy of risk prediction scores for future hospitalization and high health care costs, calculated by the Adjusted Clinical Group (ACG)'s risk stratification models, using Dutch primary health care data registries. In addition, they aimed to adjust the US-based predictive models for Dutch primary care. The statistical validity of the existing models was assessed. In addition, the underlying prediction models were trained on 95,262 patients' data from de Zoetermeer region and externally validated on data of 48,780 patients from Zeist, Nijkerk, and Urk. Information on age, sex, number of general practitioner visits, International Classification of Primary Care coded information on the diagnosis and Anatomical Therapeutic Chemical Classification coded information on the prescribed medications, were incorporated in the model. C-statistics were used to validate the discriminatory ability of the models. Calibrating ability was assessed by visual inspection of calibration plots. Adjustment of the hospitalization model based on Dutch data improved C-statistics from 0.69 to 0.75, whereas adjustment of the high-cost model improved C-statistics from 0.78 to 0.85, indicating good discrimination of the models. The models also showed good calibration. In conclusion, the local adjustments of the ACG prediction models show great potential for use in Dutch primary care.

A nationwide analysis of reperfusion therapies for pulmonary embolism in older patients with frailty.

Reperfusion therapy is challenging in the elderly. Catheter-directed therapies are an alternative for higher-risk pulmonary embolism (PE) patients if systemic thrombolysis (ST) is contraindicated or has failed. Their safety has not been evaluated in specific vulnerable populations. We aimed to assess the safety of reperfusion therapies in elderly and frail patients in the real world. In the US Nationwide Inpatient Sample from 2016 to 2020, we identified hospitalisations of patients ≥65 years with PE and defined a frailty subgroup using the Johns Hopkins Adjusted Clinical Groups frailty-defining diagnosis indicator. We investigated reperfusion therapies (ST, catheter-directed thrombolysis [CDT], catheter-based thrombectomy [CBT], surgical embolectomy [SE]) and their associated safety outcomes (overall and major bleeding). Among 980,245 hospitalisations of patients ≥65 years with PE (28.0% frail), reperfusion therapies were used in 4.9% (17.6% among high-risk PE). ST utilisation remained stable, while the use of catheter-directed therapies increased from 1.7% in 2016 to 3.2% in 2020. Among all hospitalisations with reperfusion, CDT, compared to ST, was associated with reduced major bleeding (5.8% vs 12.2%, odds ratio [OR] 0.58, 95% confidence interval [CI]: 0.49-0.70); these results also applied to frail patients. CBT, compared to SE, was also associated with reduced major bleeding (11.0% vs 22.4%, OR 0.63, 95% CI: 0.43-0.91), but not among frail patients. These differences were particularly significant in patients with non-high-risk PE. Differences persisted for overall bleeding as well. Catheter-directed therapies may be a safer alternative to classical reperfusion therapies for elderly and frail patients with PE requiring reperfusion treatment.

Developing Mixed-Effects Models to Compare the Predictive Ability of Various Comorbidity Indices in a Contemporary Cohort of Patients Undergoing Lumbar Fusion.

As incidence of operative spinal pathology continues to grow, so do the rates of lumbar spinal fusion procedures. Comorbidity indices can be used preoperatively to predict potential complications. However, there is a paucity of research defining the optimal comorbidity indices in patients undergoing spinal fusion surgery. We aimed to use modeling strategies to evaluate the predictive validity of various comorbidity indices and combinations thereof. Patients who underwent spinal fusion were queried using data from the Nationwide Readmissions Database for the years 2016 through 2019. Using comorbidity indices as predictor variables, receiver operating characteristic curves were developed for pertinent complications such as mortality, nonroutine discharge, top-quartile cost, top-quartile length of stay, and 30-day readmission. A total of 750 183 patients were included. Nonroutine discharges occurred in 161 077 (21.5%) patients. The adjusted all-payer cost for the procedure was $37 616.97 ± $27 408.86 (top quartile: $45 409.20), and the length of stay was 4.1 ± 4.4 days (top quartile: 8.1 days). By comparing receiver operating characteristics of various models, it was found that models using Frailty + Elixhauser Comorbidity Index (ECI) as the primary predictor performed better than other models with statistically significant P -values on post hoc testing. However, for prediction of mortality, the model using Frailty + ECI was not better than the model using ECI alone ( P = .23), and for prediction of all-payer cost, the ECI model outperformed the models using frailty alone ( P < .0001) and the model using Frailty + ECI ( P < .0001). This investigation is the first to use big data and modeling strategies to delineate the relative predictive utility of the ECI and Johns Hopkins Adjusted Clinical Groups comorbidity indices for the prognostication of patients undergoing lumbar fusion surgery. With the knowledge gained from our models, spine surgeons, payers, and hospitals may be able to identify vulnerable patients more effectively within their practice who may require a higher degree of resource utilization.

Association of frailty with postoperative outcomes following thoracic transplantation: A national analysis

ObjectiveFrailty has been repeatedly associated with inferior outcomes after surgical hospitalizations. However, a thorough evaluation of the impact of frailty on the clinical and financial outcomes of patients undergoing solid-organ thoracic transplantation is sparse in the literature. We evaluated the association of frailty, as determined by an administrative tool, with postoperative outcomes and healthcare resource use after heart or lung transplantation. MethodsThe Nationwide Readmissions Database was used to identify all adult hospitalizations for heart or lung transplant from 2014 to 2020. Patients were grouped as frail or nonfrail using International Classification of Diseases codes associated with conditions in the Johns Hopkins Adjusted Clinical Groups cluster. Multivariable regression models were developed to evaluate the association of frailty status on in-hospital mortality, complications, length of stay, costs, and unplanned readmissions. ResultsOf an estimated 35,862 heart or lung transplant recipients, 7316 (20.4%) were considered frail. After multivariable adjustment, frailty in heart transplantation was associated with greater odds of in-hospital mortality (adjusted odds ratio, 1.54; 95% CI, 1.19-1.99) and infectious complications (adjusted odds ratio, 1.77; 95% CI, 1.45-2.15; P < .001). Frailty in lung transplantation was also associated with higher odds of in-hospital mortality (adjusted odds ratio, 1.38; 95% CI, 1.11-1.69) and infectious complications (adjusted odds ratio, 1.93; 95% CI, 1.60-2.31). In addition, frailty in both heart transplantation and lung transplantation was associated with increased postoperative length of stay and greater costs. ConclusionsAmong transplant recipients, those classified as frail were associated with increased in-hospital mortality, perioperative complications, and resource use.

Association between Johns Hopkins Adjusted Clinical Groups risk scores and self-reported outcome measures: an observational study among individuals with complex or long-term conditions in Norway

ObjectiveInvestigate the association between Johns Hopkins Adjusted Clinical Groups (ACG) risk scores and low scores in self-reported outcome measures (SROMs) among individuals with complex or long-term conditions.DesignLongitudinal study using five...

Development and validation of predictive models for unplanned hospitalization in the Basque Country: analyzing the variability of non-deterministic algorithms

BackgroundThe progressive ageing in developed countries entails an increase in multimorbidity. Population-wide predictive models for adverse health outcomes are crucial to address these growing healthcare needs. The main objective of this study is to develop and validate a population-based prognostic model to predict the probability of unplanned hospitalization in the Basque Country, through comparing the performance of a logistic regression model and three families of machine learning models.MethodsUsing age, sex, diagnoses and drug prescriptions previously transformed by the Johns Hopkins Adjusted Clinical Groups (ACG) System, we predict the probability of unplanned hospitalization in the Basque Country (2.2 million inhabitants) using several techniques. When dealing with non-deterministic algorithms, comparing a single model per technique is not enough to choose the best approach. Thus, we conduct 40 experiments per family of models - Random Forest, Gradient Boosting Decision Trees and Multilayer Perceptrons - and compare them to Logistic Regression. Models’ performance are compared both population-wide and for the 20,000 patients with the highest predicted probabilities, as a hypothetical high-risk group to intervene on.ResultsThe best-performing technique is Multilayer Perceptron, followed by Gradient Boosting Decision Trees, Logistic Regression and Random Forest. Multilayer Perceptrons also have the lowest variability, around an order of magnitude less than Random Forests. Median area under the ROC curve, average precision and positive predictive value range from 0.789 to 0.802, 0.237 to 0.257 and 0.485 to 0.511, respectively. For Brier Score the median values are 0.048 for all techniques. There is some overlap between the algorithms. For instance, Gradient Boosting Decision Trees perform better than Logistic Regression more than 75% of the time, but not always.ConclusionsAll models have good global performance. The only family that is consistently superior to Logistic Regression is Multilayer Perceptron, showing a very reliable performance with the lowest variability.

Developing Predictive Models to Anticipate Shunt Complications in 33,248 Pediatric Patients with Shunted Hydrocephalus Utilizing Machine Learning

Introduction: Hydrocephalus is a common pediatric neurosurgical pathology, typically treated with a ventricular shunt, yet approximately 30% of patients experience shunt failure within the first year after surgery. As a result, the objective of the present study was to validate a predictive model of pediatric shunt complications with data retrieved from the Healthcare Cost and Utilization Project (HCUP) National Readmissions Database (NRD). Methods: The HCUP NRD was queried from 2016 to 2017 for pediatric patients undergoing shunt placement using ICD-10 codes. Comorbidities present upon initial admission resulting in shunt placement, Johns Hopkins Adjusted Clinical Groups (JHACG) frailty-defining criteria, and Major Diagnostic Category (MDC) at admission classifications were obtained. The database was divided into training (n = 19,948), validation (n = 6,650), and testing (n = 6,650) datasets. Multivariable analysis was performed to identify significant predictors of shunt complications which were used to develop logistic regression models. Post hoc receiver operating characteristic (ROC) curves were created. Results: A total of 33,248 pediatric patients aged 6.9 ± 5.7 years were included. Number of diagnoses during primary admission (OR: 1.05, 95% CI: 1.04–1.07) and initial neurological admission diagnoses (OR: 3.83, 95% CI: 3.33–4.42) positively correlated with shunt complications. Female sex (OR: 0.87, 95% CI: 0.76–0.99) and elective admissions (OR: 0.62, 95% CI: 0.53–0.72) negatively correlated with shunt complications. ROC curve for the regression model utilizing all significant predictors of readmission demonstrated area under the curve of 0.733, suggesting these factors are possible predictors of shunt complications in pediatric hydrocephalus. Conclusion: Efficacious and safe treatment of pediatric hydrocephalus is of paramount importance. Our machine learning algorithm delineated possible variables predictive of shunt complications with good predictive value.

Effect of Frailty on Hospital Outcomes Among Pediatric Cancer Patients in the United States: Results From the National Inpatient Sample.

Studies on frailty among pediatric patients with cancer are scarce. In this study, we sought to understand the effects of frailty on hospital outcomes in pediatric patients with cancer. This retrospective study used data collected and stored in the Nationwide Inpatient Sample (NIS) between 2005 and 2014. These were hospitalized patients and hence represented the sickest group of patients. Frailty was measured using the frailty definition diagnostic indicator by Johns Hopkins Adjusted Clinical Groups. Of 187,835 pediatric cancer hospitalizations included in this analysis, 11,497 (6.1%) were frail. The average hospitalization costs were $86,910 among frail and $40,358 for nonfrail patients. In propensity score matching analysis, the odds of in-hospital mortality (odds ratio, 2.08; 95% CI, 1.71-2.52) and length of stay (odds ratio, 3.76; 95% CI, 3.46-4.09) were significantly greater for frail patients. The findings of our study suggest that frailty is a crucial clinical factor to be considered when treating pediatric cancer patients in a hospital setting. These findings highlight the need for further research on frailty-based risk stratification and individualized interventions that could improve outcomes in frail pediatric cancer patients. The adaptation and validation of a frailty-defining diagnostic tool in the pediatric population is a high priority in the field.

Disparities in Diabetes Care: Differences Between Rural and Urban Patients Within a Large Health System.

We sought to ascertain factors associated with the quality of diabetes care, comparing rural vs urban diabetic patients in a large health care system. We conducted a retrospective cohort study assessing patients' attainment of the D5 metric, a diabetes care metric having 5 components (no tobacco use, glycated hemoglobin [A1c] level less than 8%, blood pressure less than 140/90 mm Hg, low-density lipoprotein cholesterol level at goal or statin prescribed, and aspirin use consistent with clinical recommendations). Covariates included age, sex, race, adjusted clinical group (ACG) score as a marker of complexity, insurance type, primary care clinician type, and health care use data. The study cohort consisted of 45,279 patients with diabetes, 54.4% of whom resided in rural locations. The D5 composite metric was met in 39.9% of rural patients and 43.2% of urban patients (P <.001). Rural patients were significantly less likely to have attained all metric goals than urban counterparts (adjusted odds ratio [AOR] = 0.93; 95% CI, 0.88-0.97). The rural group had fewer outpatient visits (mean number of visits = 3.2 vs 3.9, P <.001) and less often had an endocrinology visit (5.5% vs 9.3%, P <.001) during the 1-year study period. Patients with an endocrinology visit were less likely to have met the D5 metric (AOR = 0.80; 95% CI, 0.73-0.86), whereas the more outpatient visits patients had, the greater their likelihood of attainment (AOR per visit = 1.03; 95% CI, 1.03-1.04). Rural patients had worse diabetes quality outcomes than their urban counterparts, even after adjustment for other contributing factors and despite being part of the same integrated health system. Lower visit frequency and less specialty involvement in the rural setting are possible contributing factors.

D-20 | Impact of Frailty on Outcomes of Percutaneous Coronary Intervention: Insights from the Nationwide Inpatient Sample

Patients with significant coronary artery disease require percutaneous coronary intervention (PCI). However, whether frailty has adverse outcomes in patients undergoing PCI has been controversial. We investigated the impact of frailty on the outcomes of patients who underwent PCI. We reviewed the National Inpatient Sample to identify patients that underwent PCI. Frailty was defined using the Johns Hopkins Adjusted Clinical Groups frailty-defining diagnoses indicator. Univariate and multivariate logistic regression analyses were performed to determine any association between frailty and the clinical outcomes. From 2017-2020, 102,324 patients underwent PCI in the United States, and 7.3% met the criteria for frailty. We found that the frail group had longer lengths of stay (10 vs 4 days, p < 0.001), and higher hospitalization costs ($266k vs $158k, p < 0.001). After multivariate adjustment, we found that frailty was associated with increased in-hospital mortality, cardiogenic shock, increased vasopressor requirements, higher need for mechanical circulatory support, higher need for mechanical ventilation, increased incidence of post-operative stroke and post-operative sepsis, and increased risk of sudden cardiac death. The presence of frailty in patients who underwent PCI was associated with significantly higher in-hospital mortality, longer lengths of stay, higher hospitalization costs, and worsening secondary outcomes. These findings may suggest the need for an alternative approach for frail patients who are indicated for revascularization.

Identifying individuals with complex and long-term health-care needs using the Johns Hopkins Adjusted Clinical Groups System: A comparison of data from primary and specialist health care.

This study aimed to present the Johns Hopkins Adjusted Clinical Groups (ACG) System risk stratification profile of a total adult population of somatic health-care users when using data from either general practitioners (GPs) or hospital services and to compare the number and characteristics of individuals identified as having complex and long-term health-care needs in each data source. This was a registry-based study that included all adult residents (N=168,285) in four municipalities in Central Norway who received somatic health care during 2013. Risk profiles were generated using the ACG System based on age, sex and diagnoses registered by GPs or the local hospital. ACG output variables on number of chronic conditions, frailty and concurrent resource utilisation were chosen as indicators of complexity. Nearly nine out of 10 (83.9%) of the population had been in contact with a GP, and 35.4% with the hospital. The mean number of diagnoses (3.0) was equal in both sources. A larger proportion of the population had higher risk scores in all variables except frailty when comparing hospital data to GP data. This was also found when comparing individuals identified as having complex and long-term health-care needs. A similar proportion of the population was found to have complex and long-term health-care needs (hospital 6.7%, GP 6.3%), but only one in five (21.5%) were identified in both data sets. As data from GPs and hospitals identified mostly different individuals with complex and long-term health-care needs, combining data sources is likely to be the best option for identifying those most in need of special attention.