Institutional Variation in Quality of Cardiovascular Implantable Electronic Device Implantation

Cardiovascular implantable electronic devices (CIEDs), including permanent pacemakers (PPMs) and implantable cardioverter-defibrillators (ICDs) with or without cardiac resynchronization therapy (CRT) capability, are among the most common and costly devices implanted in hospitals. Although effective for treating heart rhythm abnormalities, CIEDs are associated with early device-related complications, such as infection, pneumothorax, and lead dislodgement (17). These problems may cause substantial harm to patients and may lead to additional invasive treatments, such as reoperation (17). Moreover, they increase the length of stay and are expensive to treat, with in-hospital complications costing an estimated $5000 to $20000 (3). Reducing CIED complications therefore is highly desirable to minimize avoidable patient harm and health care costs. Despite the widespread use of CIEDs, little is known about differences in CIED complications among institutions, which may suggest discrepancies in care quality. Recent data from the U.S. National Cardiovascular Data Registry show that ICD complication rates vary among hospitals, suggesting that meaningful differences in patient outcomes may result from variations in quality (8, 9). These findings have called for the implementation of ICD quality measures to reduce complications (911). However, institutional variation has not been examined outside the ICD Registry, and whether it exists in other populations and health systems is uncertain. Furthermore, although PPMs are implanted more often than ICDs, institutional variation in PPM complication rates has not been assessed systematically. Thus, whether PPM complication rates differ meaningfully among institutions is unclear. The paucity of data may explain the current lack of recommendations or quality measures for PPM complications (10, 11). Moreover, understanding institutional variation in complication rates is critical to facilitate clinical and policy efforts to inform patients and improve CIED care quality. In this study, we used population-wide data from hospitals in Australia and New Zealand to determine whether complication rates after new CIED implantation vary meaningfully among institutions. We specifically compared complication rates after PPM implantation with those after ICD placement to determine whether the frequency and type of complications and institutional variation in complication rates were similar for both procedures. Methods Data Source We used hospitalization data from the Admitted Patient Data Collection for each Australian state and territory as well as the New Zealand National Minimum Dataset (Hospital Events). These databases record encounters for all inpatient and day-only hospitalizations from all public and most private hospitals, regardless of patient age or payer. For each encounter, a standardized set of variables is collected, including patient demographic characteristics, admission and discharge dates, status of the admission (urgent or elective), primary and secondary diagnoses, all procedures, and patient status at discharge. In both countries, diagnoses and procedures are coded according to the International Classification of Diseases, 10th Revision, Australian Modification, and the Australian Classification of Health Interventions (ACHI). Previous studies showed that the accuracy of coding for cardiovascular diagnoses and procedures is greater than 85% (12). Hospitalization data were available from New Zealand (100% of the population) and 7 of the 8 Australian states and territories (99% of the population). Data were unavailable from the Northern Territory. Each patient's hospitalization was linked to subsequent hospitalizations to track readmissions and to each region's Registry of Births, Deaths and Marriages to assess postdischarge deaths. In Australia, several patient identifiers are used to link records through probabilistic matching by data linkage units in each region, with reported accuracy exceeding 99% (13). In New Zealand, hospital encounters are linked nationally with a unique patient identifier, and all deaths are recorded in the National Minimum Dataset (Hospital Events). The Human Research Ethics Committee of each Australian state and territory provided ethical approval to undertake the study. Deidentified data from New Zealand were obtained under a data use agreement with the New Zealand Ministry of Health. Study Cohort We included patients older than 18 years who received a new CIED urgently or as a planned (elective) procedure, defined by ACHI codes 38353-00 (insertion of cardiac pacemaker generator) and 38393-00 (insertion of cardiac defibrillator generator) in 2010 to 2015. Generator codes do not distinguish devices as single chamber, dual chamber, or biventricular (CRT). However, we identified CIEDs with CRT capability by the presence of a left ventricular lead implant (ACHI codes 38368-00 and 38390-01). Patients were excluded if they had a replacement CIED to ensure that the index implantation did not result from a previous complication; were having catheter ablation or cardiac surgery during the same hospitalization, because early complications may reflect the outcomes of the ablation or surgery rather than the CIED; were discharged against medical advice; or lacked the 90 days of follow-up necessary to assess complications adequately. Supplement Table 1 lists codes used to define the inclusion and exclusion criteria. Supplement. Supplementary Appendix Outcomes The primary outcome was defined as major device-related complications occurring during hospitalization or within 90 days of discharge. In-hospital complications were defined as all-cause death, device-related reoperation (lead, generator, or pocket reoperation; drainage of hematoma or abscess; and pericardial or pleural drainage), postprocedural shock, and infective endocarditis. Postdischarge complications were defined as all-cause death within 30 days; device-related reoperation (as defined in the previous sentence) up to 90 days after discharge; and rehospitalization within 90 days for a device-related complication, including a mechanical complication, infection (device infection, endocarditis, or systemic infection), a complication related to perforation or inflammation (such as a pneumothorax or pericardial effusion), a pocket-related complication (such as wound dehiscence), and venous obstruction or thromboembolism. Minor complications, such as a small pneumothorax not requiring drainage, and complications that did not warrant rehospitalization were not included. We assessed outcomes up to 90 days after discharge, because most implant-related complications occur during this period (7, 14). We counted only deaths within the first 30 days, because procedure-related deaths occur early and counting deaths beyond 30 days may capture those unrelated to the procedure. These definitions of early complications are consistent with those of earlier studies (7, 14). Supplement Tables 2 to 4 list the codes used to define these outcomes. Statistical Analysis Data are summarized as frequencies and percentages for categorical variables. Continuous variables are presented as mean plus or minus SD or as median and interquartile range (IQR). In estimating the frequency of outcomes, we counted multiple events (such as lead reoperations) for the same patient only once. We estimated each hospital's risk-standardized complication rate (RSCR) by using a hierarchical generalized linear model to account for differences in hospital case mix, sample size, and clustering of patients within hospitals by utilizing methods we used previously to profile institutional variation in procedural outcomes for public reporting (9, 15, 16). To develop the risk adjustment model, we identified patient factors independently associated with the outcome of complications by fitting a logistic regression model using a 50% random sample of the study cohort. Candidate variables consisted of those that we could extract from administrative data and that were reported in the literature as having a documented or plausible relationship to complications. These variables included age, sex, procedure status (urgent or elective), CIED type (PPM or ICD), CRT capability of the CIED, and comorbid conditions, as outlined in Supplement Table 5. Comorbid conditions were derived by using the condition category classification that groups International Classification of Diseases codes into 180 clinically meaningful conditions and using secondary diagnosis codes from the index admission and hospitalizations in the preceding 12 months (17). We purposely omitted variables that may affect quality, such as length of stay and race. We then removed the nonsignificant factors stepwise until the final model contained characteristics independently associated with the outcome at a P value less than 0.050. Next, we evaluated whether the dropped variables confounded the relationships of the significant variables to the outcome by reentering them into the model, and evaluated for clinically relevant interactions (18). None of the dropped variables led to a large (>25%) change in the estimates of the significant variables when reentered, and there were no interaction terms. We validated the model in the remaining 50% sample of the cohort. We then used the hierarchical generalized linear model to estimate a random intercept term reflecting each hospital's contribution to the risk for the outcome on the basis of its actual complication rate, the performance of other hospitals with a similar case mix, and the hospital's sample size. The RSCR is the ratio of the predicted complication rate divided by the expected complication rate multiplied by the cohort average complication rate. The predicted complication rate was calculated on the basis of each hospital's case mix and the estimated hospital-specific intercept term. The expected complication rate was calculated on the basis of eac