Predicting Persistently High Primary Care Use James M. Naessens, MPH Macaran A. Baird, MD, MS Holly K. Van Houten, BA David J. Vanness, PhD Claudia R. Campbell, PhD © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Identification of Costly Patients • Many factors related to high use • Patient demographics • Certain diagnoses • Chronic conditions • Disability • Severity of disease • Prior use (health care and medications) © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Focus of Identification • Total health care spending • Case management • Hospitalization • Disease management © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Physician Visits Employee Health Plan, 1997 50% 44% 40% 30% 20% 11% 10% 0% 0 2 4 6 # of visits Patients 8 10+ Visits © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Physician Visits - Specialty Care 50% 36% 40% 30% 20% 10% 4% 0% 0 1 2 3 4 5 6 # of visits Patients 7 8 9 10+ Visits © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Physician Visits - Primary Care 50% 40% 30% 18% 20% 10% 2% 0% 0 2 6 4 # of visits Patients 8 10+ Visits © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Reactions • Expect a small number of individuals to have a large number of visits to specialists; however, we did not expect such concentration of visits to primary care providers © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Persistence of High Primary Care Use 1997 10+ 1998 <10 PC visits 1998 10+ PC visits Pediatrics (n=152) 77.0% 23.0% Adult (n=867) 82.2% 17.8% © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. High Primary Care Use • A large percentage of primary care use may be incurred by patients seeking help on nonmedical issues (Lundin, 2001; Sweden) © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Dr. Baird’s Questions Do we have people who are “over-serviced”, but “under-served”? Can we predict who they might be (and possibly intervene)? © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Study Population • 54,074 eligible patients with research authorization • 6% of population excluded due to HIPAA and Minnesota regulations • Outpatient office visits: 1997-1999 • Primary care: • Family medicine • General internal medicine • General pediatrics • Obstetrics © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Methods • Identify factors associated with “persistent, high” primary care use: • 10+ visits in two consecutive years • Develop logistic model on 1997-1998 data • Confirm model on 1998-1999 data © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. High Users 1997 n = 929 High Users 1998 Recurrent high use in 1998 n=163 Not eligible in 1998 n = 58 New high users in 1998 n = 947 n = 987 Not eligible in 1999 n = 10 n = 1120 © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Potential Risk Factors • Age • Gender • Diagnoses • Employee/dependent status • (During timeframe: no copays, deductibles) © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Clinical Risk Factors • Adjusted Clinical Groups – Johns Hopkins • Based on all diagnoses for patient in year • Clinically meaningful • Developed by medical experts in primary care • Predictive of utilization and resource costs © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Going from Diagnosis Codes to ACGs Diagnosis Codes Adjusted Diagnosis Groups (ADGs): 32 Age, Gender (ACGs)-Adjusted Clinical Groups ©1998 © The Johns Hopkins University School of Hygiene andrights Public Health 2004 – Mayo College of Medicine, Mayo Clinic. All reserved. Illustrative ACG Decision Tree Entire Population ACG X ACG Y Assignment is based on age, gender, ADGs, and optionally, delivery status and birthweight ACG Z There are actually around 106 ACGs ©1998 The University School Mayo of Hygiene © Johns 2004 –Hopkins Mayo College of Medicine, Clinic.and AllPublic rights Health reserved. • To better understand what factors may be important in predicting primary care visits, we used the ADGs as our clinical risk factor © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Model Results – Overall: Development 33 ADG Pregnancy Odds Ratio 0.17 95% CI (0.10,0.28) Score -4 11 Chronic Med: Unstable 2.07 (1.37,3.12) +2 30 See and Reassure 2.06 (1.24,3.41) +2 26 Signs & Sympt: Minor 1.51 (1.02,2.22) +1 23 Psychosocial: Time Limited, Minor 1.56 (1.01,2.41) +1 © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. % Persistent Persistent High Primary Care Use by Model Score 60 50 40 30 20 10 0 Overall Adults Peds -4 to 2 -1 0 1 2 3 4 5+ Score © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Yield of Model Score - Adults • Using a score of 1 or greater • Sensitivity – 80.3% Specificity – 62.7% • Using a score of 2 or greater • Sensitivity – 50.3% Specificity – 81.2% • Area under ROC curve – 0.794 © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Yield of Model Score - Pediatrics Prediction among pediatrics is not useful: • score of 1 or greater • Sensitivity - 78.3% Specificity - 29.9% • score of 2 or greater • Sensitivity - 33.3% Specificity - 75.1% © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Persistence of High Primary Care Use – Confirmatory Sample 1998 10+ 1999 <10 PC visits 1999 10+ PC visits Pediatrics (n=237) 74.7% 25.3% Adult (n=873) 80.4% 19.6% © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. % Persistent Comparison of Model Scores 1998 vs 1999 70 60 50 40 30 20 10 0 1998 1999 -4 to 2 -1 0 1 2 3 4 5+ Score © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Yield of Model Score – Adults Confirmatory Data • Using a score of 1 or greater • Sensitivity – 75.8% Specificity – 57.9% • Using a score of 2 or greater • Sensitivity – 49.8% Specificity – 80.0% • Area under ROC curve – 0.752 • New persistent – 0.713 Recurrent – 0.594 © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Discussion • Unstable chronic medical conditions were predictive of continued high use. • Good candidates for disease management. © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Discussion 2 • Time-limited minor psychosocial conditions, minor signs and symptoms, and see and reassure conditions were also predictive. • These “over-serviced, under-served” may benefit from alternative social support services or integrated consultations with primary care providers to better address patient needs through non-medical approaches. © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Discussion 3 • Scoring model was able to consistently identify a sizeable portion of the persistent high users, but not effective among pediatric patients. © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Limitations • Single group of covered employees and dependents in small urban setting in a Midwestern state. • Fee-for-service coverage with no co-payments, co-insurance or deductibles at time of study. • Limited risk factors considered. © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved. Further Research • Family Practice team is evaluating “reflective interviews” and integrated consultations among patients with high primary care use. • Need to evaluate cost effectiveness of proposed interventions. © 2004 – Mayo College of Medicine, Mayo Clinic. All rights reserved.