All four physician claims-based case definitions assessed resulted in "substantial" agreement with our reference standard registry definition for chronic dialysis. One outpatient claim for dialysis was the most sensitive definition, while more complicated definitions exhibited modest increases in positive predictive value. The optimal administrative data definition may vary with the research objective. For example, when seeking to maximize identification of dialysis as an outcome an approach based on at least 1 outpatient claim may be preferable. In contrast, when establishing a cohort of patients with ESRD receiving chronic dialysis that includes the fewest non-diseased cases being captured, the use of continuous outpatient claims may be better suited.
Some of the discrepancies between our registry and physician claims algorithms for chronic dialysis likely relate to differences in the classification of patients who receive temporary dialysis or who die soon after initiating dialysis Traditionally, administrative algorithms and national registries, such as the USRDS, have required a 90-day timeframe to define chronic dialysis [19
]. Although this approach avoids identification of patients who receive temporary dialysis then recover renal function within 3 months, it introduces survivor bias and does not capture chronic dialysis patients that may begin dialysis but die before meeting the inclusion criteria of the definition. Our study demonstrates that approaches based on 1 or 2 outpatient dialysis claims are substantially more sensitive than definitions based on 90 days of claims, although this definition may include some patients who would not be classified as receiving chronic dialysis in a registry (false positive cases). Utilizing a definition that does not require the patient to survive a certain amount of time eliminates any potential survival bias and allows studies of the patient group that begin dialysis and die soon after. However the limitation of this definition is that it may also include patients with acute kidney injury requiring dialysis for a short period who subsequently recover their renal function and no longer require dialysis. Furthermore, estimates of disease incidence and outcomes will not be comparable to studies based on most existing national registries.
Establishing the validity of an outpatient administrative data definition for chronic dialysis will allow researchers to utilize physician billing claims data to assess outcomes and form cohorts. This is of international relevance, even in countries where established dialysis registries are available. In the United States, not all researchers have the means to access the USRDS. In other registries from other countries often only cross-sectional, regional data with limited outcomes are available. Thus, validated methods for identifying chronic dialysis patients using billing claims data would be useful for in health services research.
We found that the use of physician claims data resulted in the classification of patients as receiving dialysis who were not identified as such in our registry (false positives). Most of these patients were removed from the case definition when algorithms which required claims to span 90 days were used. This is in-keeping with the hypothesis that these events may be acute kidney injury cases or patients who were initiated on dialysis but subsequently recovered renal function; i.e., those not considered chronic dialysis patients and thus not captured in the registry. We also found that physician claims failed to identify some patients captured in the registry (false negatives). As Alberta Health and Wellness does not employ any formal quality assurance or correction process, this may be due to missed billings, billing errors, billings made by physicians on alternative payment plans (shadow billing) or miscoding present in administrative data sources, as the number of such patients decreased when algorithms that required less intensive physician claims were employed.
To our knowledge, this is the first study to look at using outpatient administrative data sources using procedure codes to define chronic dialysis. Others have developed algorithms for acute kidney injury and chronic kidney disease using inpatient administrative data [5
]. Given that the majority of chronic dialysis patients are treated in the outpatient setting, administrative data algorithms limited to inpatient encounters are likely to perform poorly when compared against a reference standard. Three previous studies have included outpatient claim data [14
]. However, Kern et al. excluded chronic dialysis patients, focusing on the validity of administrative data to define chronic kidney disease defined by eGFR <60 ml/min/1.73 m2
]. Neither Weintraub et al. nor Wilchesky et al. included procedural codes [14
]. Their work was limited to ICD-9-CM diagnosis codes for chronic renal failure. Thus, our study is novel, and could facilitate further health services research in a high risk population with ESRD who experience very high morbidity, mortality, and health care costs.
Our study does have several limitations. First, the billing codes used are from the Canadian Classification of Diagnostic, Therapeutic and Surgical Procedures (CCP); a classification system developed and applied in Canada. However, most countries have similar billing practices and billing codes that could be mapped to the CCP codes. Second, we used a provincial registry of all chronic dialysis patients as the reference standard. Although this registry is geographically inclusive, some dialysis patients may be omitted from the registry in error, thereby resulting in misclassification. However, as this registry is linked to ongoing dialysis treatment, the number of patients not registered is expected to be small. Third, our study did not distinguish between dialysis modalities (hemodialysis versus peritoneal dialysis, or in-centre versus home dialysis), and the accuracy of patient registry and physician claims in these settings may vary. However, prior research has reported limitations in the accuracy of administrative data for identifying the timing of changes between dialysis modalities suggesting that administrative data sources may be better suited to the general identification of patients receiving chronic dialysis rather than a specific modality [29