Can Dialysis Patients Be Accurately Identified Using Healthcare Claims Data?

Abstract

♦ Background: While health insurance claims data are often used to estimate the costs of renal replacement therapy in patients with end-stage renal disease (ESRD), the accuracy of methods used to identify patients receiving dialysis — especially peritoneal dialysis (PD) and hemodialysis (HD) — in these data is unknown.

♦ Methods: The study population consisted of all persons aged 18 - 63 years in a large US integrated health plan with ESRD and dialysis-related billing codes (i.e., diagnosis, procedures) on healthcare encounters between January 1, 2005, and December 31, 2008. Using billing codes for all healthcare encounters within 30 days of each patient’s first dialysis-related claim (“index encounter”), we attempted to designate each study subject as either a “PD patient” or “HD patient.” Using alternative windows of ± 30 days, ± 90 days, and ± 180 days around the index encounter, we reviewed patients’ medical records to determine the dialysis modality actually received. We calculated the positive predictive value (PPV) for each dialysis-related billing code, using information in patients’ medical records as the “gold standard.”

♦ Results: We identified a total of 233 patients with evidence of ESRD and receipt of dialysis in healthcare claims data. Based on examination of billing codes, 43 and 173 study subjects were designated PD patients and HD patients, respectively (14 patients had evidence of PD and HD, and modality could not be ascertained for 31 patients). The PPV of codes used to identify PD patients was low based on a ± 30-day medical record review window (34.9%), and increased with use of ± 90-day and ± 180-day windows (both 67.4%). The PPV for codes used to identify HD patients was uniformly high — 86.7% based on ± 30-day review, 90.8% based on ± 90-day review, and 93.1% based on ± 180-day review.

♦ Conclusions: While HD patients could be accurately identified using billing codes in healthcare claims data, case identification was much more problematic for patients receiving PD.

Much of what we know about differences in clinical outcomes and costs of care between patients receiving hemodialysis (HD) versus peritoneal dialysis (PD) comes from analyses of Medicare claims data, since persons with end-stage renal disease (ESRD) are eligible for health insurance coverage under this federal program after the first three months of dialysis. Total annual healthcare expenditures have been reported to be significantly lower among Medicare enrollees receiving PD in comparison with those receiving HD (1,2).

In most studies to date, identification of patients who received PD versus HD has been based on various diagnosis and/or procedure codes used for third-party billing and reimbursement (2-6). To the best of our knowledge, however, the predictive accuracy of these techniques for identifying PD and HD patients in health insurance claims data is unknown. The possibility exists, therefore, that the dialysis modality that some patients actually received was incorrectly designated on claims. To shed light on this issue, we examined the predictive accuracy of various billing codes as a means of identifying patients who received PD versus HD, using information gleaned from electronic medical records (EMRs) as our “gold standard.”

Methods

Data Source

This retrospective study was conducted at Henry Ford Health System (HFHS), a comprehensive health system that provides medical care to approximately 800,000 residents of Detroit, Michigan, and the surrounding areas, with 3.2 million patient contacts annually. Health Alliance Plan (HAP) is a wholly owned, not-for-profit health maintenance organization within HFHS that provides insurance for 475,000 persons, 125,000 of whom have elected HFHS facilities as their assigned site of care. Study subjects were drawn from the population of persons enrolled in HAP with HFHS assignment, nearly 20% of whom are aged 65 years or older.

HFHS uses a comprehensive multi-dimensional EMR system that provides clinicians and researchers with real-time access to computerized medical records (“CarePlus”). CarePlus maintains information on patient demographics, ambulatory care visits, clinical laboratory and radiology results, and inpatient admissions, as well as various other clinical and economic measures. HFHS also maintains a large administrative data warehouse containing information on all encounters with HFHS providers and facilities, including ambulatory care visits (outpatient clinic, emergency department), hospital admissions, healthcare services provided at non-HFHS sites, billing records generated within inpatient and outpatient settings, and outpatient prescription claims. Billing records include information on type of service, date of service, provider name and specialty, site of service and type of encounter, and diagnoses.

Information in CarePlus, while stored electronically, is not searchable and cannot be harvested in digital format; it therefore was extracted manually onto hard-copy case-report forms that were developed for use in this study. To ensure patient confidentiality and compliance with the Health Insurance Portability and Accountability Act (HIPAA) of 1996, no patient-identifying information was extracted. Each patient in the study sample was assigned a unique study identifier, which was used to link information from different sources. The study was approved by the HFHS Institutional Review Board.

Study Sample

The source population for the study consisted of all persons, aged 18 - 63 years as of January 1, 2005, who were enrolled in HAP with HFHS assignment anytime between January 1, 2005, and December 31, 2008 (the “study period”). (In the US, patients aged 65 years or older receive insurance coverage through the Medicare program; since we did not have access to Medicare data, patients of this age were not included in our study sample. Medicare is also the program that provides health insurance to persons with ESRD aged < 65 years. However, because Medicare does not become the primary payer for ESRD patients of this age until approximately 30 months following disease onset, we limited our attention to patients between the ages of 18 and 63 years as of the date of initiation of dialysis to eliminate problems of incomplete data capture.) Among these persons, we identified all those with evidence of ESRD and dialysis-related healthcare encounters during the study period.

Encounters related to dialysis were designated principally using diagnosis and/or procedure codes for dialysis-related services rendered in either outpatient or inpatient settings (a listing of all dialysis-related codes is included in the Appendices). The date of the first dialysis-related encounter was designated the “index encounter,” and persons with less than one year of HAP enrollment following their index encounter were excluded from the study sample. Evidence of ESRD was ascertained based on presence of one or more healthcare encounters with International Classification of Diseases (ICD)-9-CM diagnosis code 585.6 during the year preceding the index encounter and/or the six-month period subsequent to it (patients were required to have a diagnosis of ESRD in the 12-month period preceding the date of the earliest claim for dialysis, and/or the 6-month period thereafter).

We classified all dialysis-related billing codes a priori as indicative of either PD or HD (Appendix A1 and A2). We then examined billing codes for all patients within 30 days of their index encounter, and designated (as feasible) each patient as receiving PD or HD; a 30-day period was used for review of billing codes, since the code for the index encounter often was not sufficiently descriptive to permit classification.

APPENDIX A1 -.

PERITONEAL DIALYSIS-RELATED PROCEDURE/DIAGNOSIS CODES

APPENDIX A2 -.

HEMODIALYSIS-RELATED PROCEDURE/DIAGNOSIS CODES

Following this designation using claims data only, trained medical abstractors reviewed each patient’s medical record to determine the dialysis modality actually received, using alternative windows of ± 30 days, ± 90 days, and ± 180 days around the index encounter.

Measures and Analyses

We examined the predictive accuracy of healthcare claims for designating patients as receiving PD versus HD, using information in the EMR as our “gold standard.” Accordingly, patients were deemed “true-positives” if review of medical records revealed evidence of the designated dialysis modality; they were deemed “false-positives” if the designated dialysis modality could not be confirmed in this fashion.

We estimated the predictive accuracy of dialysis-related billing codes for PD and HD, respectively, in healthcare claims using positive predictive value (PPV), defined as the ratio of the total number of patients who were “true-positives” to the total number of patients who were either “true-positives” or “false-positives”. Since PPV was anticipated to be dependent upon the timeframe employed for medical record review, we alternatively employed time windows of ± 30 days, ± 90 days, and ± 180 days around each patient’s index encounter (Figure 1). Ninety-five percent confidence intervals (95% CI) for PPV were estimated using a normal approximation of the binomial distribution.

Figure 1 —

Illustration of estimation of positive predictive value.

Results

We identified a total of 233 ESRD patients with evidence of dialysis-related encounters in healthcare claims data during the study period; 43 and 173 patients were designated as receiving PD and HD, respectively (14 patients had evidence of both modalities and were consequently included in both groups). Dialysis modality could not be determined for 31 patients (i.e., their billing codes were nonspecific). Most patients designated as receiving PD had healthcare encounters with current procedural terminology (CPT) code 49421 and/or 90945 (Table 1). Almost all patients designated as receiving HD had healthcare encounters with CPT codes 36145 or 90935, ICD-9-CM diagnosis code V56.0, and/or ICD-9-CM procedure codes 38.95 and 39.95.

TABLE 1.

Frequently Noted PD- and HD-related Codes

The PPV of billing codes used to identify PD patients was low (34.9%) (95% CI: 20.6%, 49.1%) based on a ± 30-day window (around the index date) for medical record review; it improved to 67.4% (53.4%, 81.4%) when the window for review was extended to either ± 90 days or ± 180 days (Table 2). The PPV of billing codes used to identify HD patients was uniformly high: 86.7% (81.6%, 91.8%) at ± 30 days, 90.8% (86.4%, 95.1%) at ± 90 days, and 93.1% (89.3%, 96.8%) at ± 180 days (Table 3). Among the most commonly encountered codes, CPT-4 code 49421 had a low PPV (40.9%) for PD in a ±30-day window, but high (95.5%) with either a ± 90-day or ± 180-day window; the corresponding CPT-4 code for HD (36145) had high PPVs for HD at ± 30 days (89.3%), ± 90 days (92.9%), and ± 180 days (96.4%). Procedure codes for dialysis-related encounters had relatively low PPVs for PD and relatively high PPVs for HD irrespective of time window. For example, CPT-4 code 90945, which was designated a priori as PD-related, had a PPV of 31.8% at ± 30 days, 45.5% at ± 90 days, and 45.5% at ± 180 days; HD-related CPT-4 code 90935 had corresponding PPVs of 95.3%, 96.9%, and 96.9%, respectively. Other frequently used HD-related codes also had high (i.e., >90%) PPVs regardless of time window employed.

TABLE 2.

PPVs of Individual PD-related Codes^a

TABLE 3.

PPVs of HD-related Codes^a

Discussion

Our study is the first, to the best of our knowledge, to examine the predictive accuracy in healthcare claims data of billing (i.e., procedure, diagnosis) codes that are often used to identify ESRD patients who receive PD and HD. Our findings indicate that while patients who are receiving HD can be identified with reasonable accuracy, identification of patients receiving PD is much more problematic. Our findings raise potential concerns about comparisons of PD and HD patients that are based on analyses of US healthcare claims data, as their findings are only as robust as the methods used to identify patients receiving these dialysis modalities.

There are two principal explanations for our findings. First, while the most frequently used codes used to identify HD patients were specific for this modality, those used to identify PD patients — CPT-4 codes 49421 and 90945 — were not. Although the PPV for the first of these codes was low (41%) using a 30-day window, it was quite high (96%) with both the 90-day and 180-day windows, suggesting that often there may be a substantial delay between catheter placement and commencement of PD, indicative of the typically planned nature of this modality (i.e., while it can be used in emergency situations (7), there typically is a lag between catheter placement and initiation of PD to allow for recovery from surgery and the development of scar tissue that helps hold the catheter in place). The delay between surgery and dialysis initiation may be problematic in terms of evaluating clinical outcomes and costs, as the date of the procedure code does not approximate the date that PD is actually initiated. The PPV of the second code (i.e., CPT-4 90945) remained low irrespective of the window used for medical record review, suggesting that it is commonly used for dialysis modalities other than PD (e.g., hemofiltration).

Second, the total number of patients with billing codes for PD was much lower than the number with such codes for HD (43 vs 173, respectively), and the number with dialysis confirmed through medical record review was even lower (29 vs 161). While both estimates yield a point-prevalence estimate for PD greater than the national average (18% and 15% vs ∼7% (1,8)), robustness of findings is compromised with a small sample size. Furthermore, because the number of PD patients was relatively low, we were unable to assess the PPV of all codes that are available for reimbursement of PD. For example, there were no medical encounters with the accompanying PD-related CPT-4 code 90947.

In addition, billing directives from insurance companies are not always aligned with services rendered during visits. For example, the PD-related CPT-4 code 90945 had a PPV of only 46% over a ± 180-day window. It is possible that this code is used to bill for predialysis services, and that some patients are readied for PD, but ultimately receive HD instead. In a recent examination of 217 US patients who initiated dialysis at a single center, 124 (57%) opted for education on PD; only 48% of these patients (i.e., 59/124) began dialysis with this modality, however (9).

Limitations of our study should be noted. The extent to which coding practices for dialysis at HFHS reflect those at other health systems across the US is unknown. In fact, there is evidence suggesting that the specific administrative codes used to bill for dialysis-related care vary substantially by payer (3). Accordingly, the degree to which codes commonly used to identify dialysis patients in our study setting are used in other settings is unknown; similarly, the PPV of codes commonly used in other health plans but not at HFHS also is unknown. Similarly, we limited our attention to patients aged < 63 years to ensure that we could construct complete chronologies of care. While we have no reason to believe that diagnostic and procedure coding on healthcare claims differs substantially across third-party payers and therefore that our results would not be generalizable to persons aged 65 years or greater, this remains a limitation of our study. Finally, our sample size was small (n = 233), particularly with respect to the number of patients identified as receiving PD (n=43), which limits the robustness of our findings. Our results therefore should be confirmed in future research.

In conclusion, our findings suggest that ESRD patients who receive HD can be accurately identified using billing codes in healthcare claims data, but that similar identification of those receiving PD is more problematic. When undertaking retrospective studies using these methods of case ascertainment, researchers should keep these problems firmly in mind.

Disclosures

Support: Funding for this study was provided by Baxter Healthcare Corporation, McGaw Park, IL. Baxter Healthcare Corporation assisted with the interpretation of data, drafting of the manuscript, and the decision to submit the manuscript for publication.

Financial Disclosure: Ms. Taneja, Mr. Berger, and Dr. Oster are employees of Policy Analysis Inc., a contract research organization that has received funding from Baxter Healthcare Corporation as well as other biomedical firms. Dr. Lamerato, Mr. Wolff, and Mr. Sheehan are employees of Henry Ford Health System. Dr. Sloand is an employee of Baxter Healthcare Corporation, the sponsor of the study, and reports owning stock in that company. Mr. Inglese is a former employee of Baxter Healthcare Corporation, and is currently employed by Hollister Incorporated.