1932. HCUP-based risk score for predicting herpes zoster
Session: Poster Abstract Session: Vaccines: Varicella/Zoster
Saturday, October 10, 2015
Room: Poster Hall
Posters
  • 1932_IDWPoster.pdf (488.4 kB)
  • Background: To control confounding in a study of the long-term effectiveness of the shingles vaccine (Zostavax™), we developed an HCUP-based herpes zoster (HZ) risk score.

    Methods: We identified 42,157 HZ cases among 1,309,280 persons ages 50 and older from 2007 to 2013, in Kaiser Permanente Northern California. We categorized all ICD-9 diagnoses from medical encounters in the year prior to HZ using the Healthcare Cost and Utilization (HCUP) 4-level Clinical Classification Software. We used all 136 second-level categories and 11 selected third-level categories.  We included 126 HCUP categories in the risk score model, dropping 21 categories with less than 0.1% person-time.  Using Cox regression with a calendar timeline, we examined HZ onset relative to each of the 126 HCUP categories during unvaccinated person-time, adjusting for age and sex.  For each person, the fitted model yielded a predicted log hazard ratio relative to one without a diagnosis in these categories, which we took as a risk score and used to adjust analyses of vaccine effectiveness (VE) in the full cohort. We compared VE adjusted by this risk score with VE adjusted by all 126 of the underlying diagnostic categories.  

    Results: Persons in the highest decile on the risk score were about 3.5 times more likely to acquire HZ than those in the lowest decile. Of the 47 HCUP categories with significantly elevated HRs, the 4 highest HRs were for cancer of lymphatic and hematopoietic tissue (2.15), HIV infection (1.62), rheumatoid arthritis and related disease (1.45), and secondary malignancies (1.40). Nine HCUP categories were associated with significantly lowered HRs.  Estimates of VE (and their confidence intervals) from the two models, using the risk score versus all 126 categories, were virtually identical. Use of the risk score, rather than 126 covariates, allowed us to reduce computation time from 20 hours to 1 hour, and facilitated examination of how VE varies across subgroups, time intervals, and level of risk of HZ.

    Conclusion: An HCUP-based risk score based on unvaccinated person-time was helpful in summarizing risk for HZ due to comorbidities in a study of vaccine effectiveness, and reduced the computational burden considerably. The score could be useful in other studies of HZ.   

    Roger Baxter, MD, FIDSA1, Joan Bartlett, MPH, MPP1, Bruce Fireman, MA1, Yong Chen, PhD, MHS2, John Hansen, MPH1, Laurie Aukes, RN1, Ned Lewis, MPH1, Patricia Saddier, MD, PhD2, M. Alan Brookhart, PhD3, Morgan Marks, PhD2, Alec Walker, MD, DrPH4 and Stefan Gravenstein, MD, MPH, CMD5, (1)Kaiser Permanente Vaccine Study Center, Oakland, CA, (2)Merck & Co., Inc. Pharmacoepidemiology Department, North Wales, PA, (3)UNC Gillings School of Global Public Health, Chapel Hill, NC, (4)World Health Information Science Consultants, LLC, Newton, MA, (5)Case Western Reserve University, Cleveland, OH

    Disclosures:

    R. Baxter, Merck: Grant Investigator , Research grant
    GSK: Grant Investigator , Research grant
    Pfizer: Grant Investigator , Research grant
    Sanofi Pasteur: Grant Investigator , Research grant

    J. Bartlett, None

    B. Fireman, None

    Y. Chen, Merck & Co., Inc.: Employee and Shareholder , Salary

    J. Hansen, None

    L. Aukes, None

    N. Lewis, None

    P. Saddier, Merck Sharp & Dohme, Corp.: Employee and Shareholder , Salary

    M. A. Brookhart, Merck: Scientific Advisor , Research support

    M. Marks, Merck Sharp & Dohme, Inc.: Employee and Shareholder , Salary

    A. Walker, Merck Research Labs: Scientific Advisor , Consulting fee

    S. Gravenstein, Merck: Scientific Advisor , Consulting fee
    Sanofi Pasteur: Consultant , Investigator , Scientific Advisor and Speaker's Bureau , Consulting fee , Grant recipient , Research support and Speaker honorarium

    Findings in the abstracts are embargoed until 12:01 a.m. PDT, Wednesday Oct. 7th with the exception of research findings presented at the IDWeek press conferences.