Currently Viewing:
The American Journal of Managed Care October 2018
Putting the Pieces Together: EHR Communication and Diabetes Patient Outcomes
Marlon P. Mundt, PhD, and Larissa I. Zakletskaia, MA
Primary Care Physician Resource Use Changes Associated With Feedback Reports
Eva Chang, PhD, MPH; Diana S.M. Buist, PhD, MPH; Matt Handley, MD; Eric Johnson, MS; Sharon Fuller, BA; Roy Pardee, JD, MA; Gabrielle Gundersen, MPH; and Robert J. Reid, MD, PhD
From the Editorial Board: Bruce W. Sherman, MD
Bruce W. Sherman, MD
Recent Study on Site of Care Has Severe Limitations
Lucio N. Gordan, MD, and Debra Patt, MD
The Authors Respond and Stand Behind Their Findings
Yamini Kalidindi, MHA; Jeah Jung, PhD; and Roger Feldman, PhD
The Characteristics of Physician Practices Joining the Early ACOs: Looking Back to Look Forward
Stephen M. Shortell, PhD, MPH, MBA; Patricia P. Ramsay, MPH; Laurence C. Baker, PhD; Michael F. Pesko, PhD; and Lawrence P. Casalino, MD, PhD
Nudging Physicians and Patients With Autopend Clinical Decision Support to Improve Diabetes Management
Laura Panattoni, PhD; Albert Chan, MD, MS; Yan Yang, PhD; Cliff Olson, MBA; and Ming Tai-Seale, PhD, MPH
Medicare Underpayment for Diabetes Prevention Program: Implications for DPP Suppliers
Amanda S. Parsons, MD; Varna Raman, MBA; Bronwyn Starr, MPH; Mark Zezza, PhD; and Colin D. Rehm, PhD
Currently Reading
Clinical Outcomes and Healthcare Use Associated With Optimal ESRD Starts
Peter W. Crooks, MD; Christopher O. Thomas, MD; Amy Compton-Phillips, MD; Wendy Leith, MS, MPH; Alvina Sundang, MBA; Yi Yvonne Zhou, PhD; and Linda Radler, MBA
CMS HCC Risk Scores and Home Health Patient Experience Measures
Hsueh-Fen Chen, PhD; J. Mick Tilford, PhD; Fei Wan, PhD; and Robert Schuldt, MA
An Early Warning Tool for Predicting at Admission the Discharge Disposition of a Hospitalized Patient
Nicholas Ballester, PhD; Pratik J. Parikh, PhD; Michael Donlin, MSN, ACNP-BC, FHM; Elizabeth K. May, MS; and Steven R. Simon, MD, MPH
Gatekeeping and Patterns of Outpatient Care Post Healthcare Reform
Michael L. Barnett, MD, MS; Zirui Song, MD, PhD; Asaf Bitton, MD, MPH; Sherri Rose, PhD; and Bruce E. Landon, MD, MBA, MSc

Clinical Outcomes and Healthcare Use Associated With Optimal ESRD Starts

Peter W. Crooks, MD; Christopher O. Thomas, MD; Amy Compton-Phillips, MD; Wendy Leith, MS, MPH; Alvina Sundang, MBA; Yi Yvonne Zhou, PhD; and Linda Radler, MBA
Optimal end-stage renal disease (ESRD) starts were associated with lower 12-month morbidity, mortality, and inpatient and outpatient utilization in an integrated healthcare delivery system.
Participants

Patients were included in the analysis if they began renal replacement therapy during the observation period, were 18 years or older on the date they started renal replacement therapy, and had continuous insurance coverage throughout the previous year. We excluded patients who recovered sufficient kidney function to stop dialysis within 3 months of initiation. Kaiser Permanente regions use nurse care coordinators to transition patients into renal replacement therapy; we compiled the patient lists of care coordinators and dialysis and transplantation authorization lists from the participating regions to identify the included population. The medical record numbers of all patients were verified, and all patients were included in the analysis.

Measures

Dependent variables. Measured outcomes included 12-month rates of sepsis and all-cause mortality. Sepsis was identified by International Classification of Diseases, Ninth Revision (ICD-9) codes (sepsis, 995.9x; septicemia, 038.x). Deaths were identified by a monthly master file provided by the Social Security Administration and filtered for the Social Security numbers of study participants. Utilization outcomes included inpatient stays, total inpatient days, emergency department (ED) visits, and outpatient visits to primary care and specialty care, which included nephrology and vascular surgery visits and visits to all other specialties. Nephrology visits did not include encounters at dialysis facilities. All data on outcomes were available through the electronic health record (EHR).

Independent variable. The primary independent variable was optimal ESRD starts, which we dichotomously measured as occurring or not occurring. Optimal starts were indicated by initial renal replacement therapy consisting of a pre-emptive kidney transplant, outpatient dialysis with peritoneal dialysis (including on an urgent basis), or outpatient dialysis with hemodialysis via an AVF or AVG, including home hemodialysis. The full measure specification is available online.14 Initiation of renal replacement therapy with hemodialysis via a CVC indicated a nonoptimal start. Data were available in Kaiser Permanente’s integrated EHR.

Covariates. Covariates included demographic and clinical characteristics. Sex, race/ethnicity, and region were measured categorically. Dichotomous variables were created for income (annual household income less than $100,000 and $100,000 or more) and education (high school degree or less and 1 or more years of college). To control for baseline utilization, we created an ordinal variable for combined hospital and ED use in the year before starting renal replacement therapy of 0 or 1 encounters, 2 to 5 encounters, and ​6 or more encounters. We created dichotomous variables for body mass index using the cut point for excess weight of 25 kg/m2, alcohol use (yes and no), and smoking status (current, former, never, and passive). Comorbidities—coronary artery disease, congestive heart failure (CHF) or fluid overload, CKD, peripheral edema, peripheral artery disease, proteinuria, diabetes, and hypertension—were measured dichotomously as present or not. We included the presence of a CKD code as a covariate because appropriate CKD coding would be expected to associate with optimal starts, and CKD codes were not always present.15 Charlson Comorbidity Index scores and glomerular filtration rate (GFR) were assessed as continuous variables.16 All data were available in the EHR; comorbidities were identified by ICD-9 codes. Data on income and education are not routinely collected during care and were imputed from block-level Census data based on participants’ home addresses.17,18 Negligible missingness of data was disregarded in analyses (Table 1 [part A and part B]).

Statistical Analysis

We calculated baseline characteristics of patients with optimal and nonoptimal starts using t tests for continuous variables and χ2 tests for categorical variables. We then created a matched data set to test for differences in outcomes between patients with optimal and nonoptimal renal replacement therapy starts. We did this by generating a propensity score for the likelihood that patients would have an optimal ESRD start using logistic regression modeling that included all listed covariates; stepwise selection was used to identify covariates significantly associated with an optimal start. The final propensity score model included GFR, education, alcohol use, coronary artery disease, CHF or fluid overload, CKD, hypertension, and peripheral edema. We matched patients with optimal starts to patients with nonoptimal starts using the greedy-5 algorithm and a matching ratio of 1:1.19,20 All optimal and nonoptimal starts were separately ordered by propensity score, and each optimal start was matched to the nearest unmatched nonoptimal start. Five-digit matches were completed first, followed by 4-digit matches, continuing down to a 1-digit match on propensity scores. Matches were not reconsidered, and unmatched optimal starts were not included in further analyses.

We calculated standardized differences in means for all covariates before and after matching, with 10% or more indicating imbalance.21 After matching, absolute standardized differences in means were less than 10% for all variables used to calculate propensity scores, except for region. Although the standardized difference in means for education was 8% (Table 1), the χ2 test for differences was statistically significant at P = .01. All subsequent analyses were performed on the propensity score–matched cohort and adjusted for propensity score, region, and education. Healthcare utilization was also adjusted for prior-year utilization.

We estimated relative rates for utilization, adjusting for prior-year utilization, propensity score, region, and education. We used logistic regression to calculate odds ratios (ORs) for sepsis and mortality and Cox proportional hazards to estimate a hazard ratio (HR) for 12-month mortality; all were adjusted for propensity score, region, and education. Analyses were performed using SAS version 9.2 (SAS Institute; Cary, North Carolina). This quality improvement analysis did not meet criteria for institutional review board oversight.


 
Copyright AJMC 2006-2019 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
x
Welcome the the new and improved AJMC.com, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up