Background

Symptomatic venous thromboembolism is a common postoperative complication. The Agency for Healthcare Research and Quality (AHRQ) has developed a Patient Safety Indicator 12 to assist hospitals, payers, and other stakeholders to identify patients who experienced this complication.

Objectives

To determine whether newly created and recently redefined ICD-9-CM codes improved the criterion validity of Patient Safety Indicator 12, based on new samples of records dated after October 2009.

Research design, subjects, measures

Two sources of data were used: (1) UHC retrospective case-control study of risk factors for acute symptomatic venous thromboembolism occurring within 90 days after total knee arthroplasty in teaching hospitals; (2) chart abstraction data by volunteer hospitals participating in the Validation Pilot Project of the AHRQ.

Results

In the UHC sample, the positive predictive value (PPV) was 99% (125/126) and the negative predictive value was 99.4% (460/463). In the AHRQ sample, the overall PPV was 81% (126/156).

Conclusions

The PPV based on both samples shows substantial improvement compared with the previously reported PPVs of 43%-48%, suggesting that changes in ICD-9-CM code architecture and better coding guidance can improve the usefulness of coded data.

Objective

To reweight the Agency for Healthcare Research and Quality Patient Safety for Selected Indicators Composite (Patient Safety Indicator [PSI] 90) from weights based solely on the frequency of component PSIs to those that incorporate excess harm reflecting patients' preferences for outcome-related health states.

Data sources

National administrative and claims data involving hospitalizations in nonfederal, nonrehabilitation, acute care hospitals.

Study design

We estimated the average excess aggregate harm associated with the occurrence of each component PSI using a cohort sample for each indicator based on denominator-eligible records. We used propensity scores to account for potential confounding in the risk models for each PSI and weighted observations to estimate the "average treatment effect in the treated" for those with the PSI event. We fit separate regression models for each harm outcome. Final PSI weights reflected both the disutilities and the frequencies of the harms.

Data collection/extraction methods

We estimated PSI frequencies from the 2012 Healthcare Cost and Utilization Project State Inpatient Databases with present on admission data and excess harms using 2012-2013 Centers for Medicare & Medicaid Services Medicare Fee-for-Service data.

Principal findings

Including harms in the weighting scheme changed individual component weights from the original frequency-based weighting. In the reweighted composite, PSIs 11 ("Postoperative Respiratory Failure"), 13 ("Postoperative Sepsis"), and 12 ("Perioperative Pulmonary Embolism or Deep Vein Thrombosis") contributed the greatest harm, with weights of 29.7%, 21.1%, and 20.4%, respectively. Regarding reliability, the overall average hospital signal-to-noise ratio for the reweighted PSI 90 was 0.7015. Regarding discrimination, among hospitals with greater than median volume, 34% had significantly better PSI 90 performance, and 41% had significantly worse performance than benchmark rates (based on percentiles).

Conclusions

Reformulation of PSI 90 with harm-based weights is feasible and results in satisfactory reliability and discrimination, with a more clinically meaningful distribution of component weights.