Original Article
Administrative data accurately identified intensive care unit admissions in Ontario

https://doi.org/10.1016/j.jclinepi.2005.11.015Get rights and content

Abstract

Background and Objectives

To evaluate the accuracy of Ontario administrative health data for identifying intensive care unit (ICU) patients.

Materials and Methods

Records from the Critical Care Research Network patient registry (CCR-Net) were linked to the Ontario Health Insurance Program (OHIP) database and the Canadian Institute for Health Information (CIHI) database. The CCR-Net was considered the criterion standard for assessing the accuracy of different OHIP or CIHI codes for identifying ICU admission.

Results

The highest positive predictive value (PPV) for ICU admission (91%) was obtained using a CIHI special care unit (SCU) code, but its sensitivity was poor (26%). A strategy based on a combination of CIHI SCU codes yielded a lower PPV (84%) but a higher sensitivity (92%). A strategy based purely on OHIP claims yielded further reductions in PPV (73%), gains in specificity (99%), and moderate sensitivity (56%). The highest sensitivity (100%) was obtained using a combination of CIHI and OHIP codes in exchange for poor PPV (32%).

Conclusions

Administrative databases can be used to identify ICU patients, but no single strategy simultaneously provided high sensitivity, specificity, and PPV. Researchers should consider the study purpose when selecting a strategy for health services research on ICU patients.

Introduction

Administrative databases are used in health services research because they offer large and comprehensive sample size, systematic and coordinated data collection, and population-based information free of referral bias. We evaluated the accuracy of the Canadian Institute of Health Information database (which contains demographic, administrative, and clinical data for all hospital discharges and day surgeries in Canada), and the Ontario Health Insurance Program database (which contains all fee-for-service claims paid for services provided to Ontario residents) for identifying intensive care unit admissions.

Admission to the Critical Care Research Network (CCR-Net) database was our criterion standard. The CCR-Net is a collaboration including 33 Ontario hospitals that maintains a patient registry of the activities of critical care facilities [1], [2], [3]. The minimal dataset of admissions is systematically collected by participating hospitals and includes patient demographics, data to calculate a score on the Acute Physiology and Chronic Health Evaluation II (APACHE II) system [4], and information about ICU and hospital length of stay. The CCR-Net data has good interobserver reliability and strong face validity [5], [6], [7].

Section snippets

Selection of hospitals and patients

We considered all admissions (according to date of arrival) to nine Ontario acute care hospitals (including one teaching hospital) participating in the CCR-Net and appearing in the Canadian Institute of Health Information (CIHI) database during January 1, 2001 to December 31, 2002. The specific hospitals were selected based on the rates of successful linkage (>90% agreement) between the CIHI and CCR-Net databases (Appendix 1).

Handling of duplicate records

The patient admission to CIHI containing the most recent discharge

Results

During the 2-year study interval a total of 329,284 admissions to CIHI with valid health care numbers occurred at the nine hospitals. After deleting 203 duplicate records, 329,081 admissions were available for analysis. During the same time period, 19,876 records appeared in CCR-Net and successful linkage to CIHI was possible for 18,637 (94%). After removing 936 records in CCR-Net representing readmissions to ICU during the same hospital admission, 17,701 ICU admissions served as the criterion

Discussion

We explored administrative data from multiple Ontario hospitals and found that the databases can identify ICU patients. Several strategies were evaluated, but no one strategy simultaneously provided ideal sensitivity, specificity, and positive predictive value. OHIP codes and CIHI SCU codes had excellent sensitivity and specificity, but their associated positive predictive values were only fair (54 and 35%, respectively). This is not surprising, given the low overall frequency (prevalence 5.4%)

Acknowledgments

All authors participated in the study design, interpretation of results, and writing of the final draft. The corresponding author had full access to all the data in the study, and had final responsibility for the decision to submit for publication. No author has a financial conflict of interest that could inappropriately bias this work. Donald A. Redelmeier is supported by the Canada Research Chair in Medical Decision Sciences.

We are indebted to Dr. Andreas Laupacis for providing helpful

References (11)

  • S.P. Keenan et al.

    The Critical Care Research Network: a partnership in community-based research and research transfer

    J Eval Clin Pract

    (2000)
  • W.J. Sibbald et al.

    The Ontario Health Care Evaluation Network and the Critical Care Research Network as vehicles for research transfer

    Med Decis Making

    (1998)
  • S.P. Keenan et al.

    Creation of a critical care research network

    Curr Opin Crit Care

    (1998)
  • W.A. Knaus et al.

    APACHE II: a severity of disease classification system

    Crit Care Med

    (1985)
  • L.M. Chen et al.

    Interobserver variability in data collection of the APACHE II score in teaching and community hospitals

    Crit Care Med

    (1999)
There are more references available in the full text version of this article.

Cited by (80)

  • COVID Vaccine Hesitancy and Long-Term Traffic Risks

    2024, American Journal of Medicine
  • Long-Term Outcomes of Cardiogenic Shock Complicating Myocardial Infarction

    2023, Journal of the American College of Cardiology
  • COVID Vaccine Hesitancy and Risk of a Traffic Crash

    2023, American Journal of Medicine
View all citing articles on Scopus
View full text