The Role of Overbilling in Hospitals’ Earnings Management Decisions

Abstract

This paper examines the role of overbilling in hospitals’ earnings management choices. Overbilling by hospitals is a form of revenue manipulation that involves misclassifying a patient into a diagnosis-related group that yields higher reimbursement. As overbilling allows hospitals to increase revenues without altering operations, affecting costs, or having to reverse such behavior in the future, I propose and find that overbilling reduces hospitals’ use of managing accruals or cutting discretionary expenditures. Next, I find that hospital managers prefer overbilling to managing accruals (cutting discretionary expenditures) when cutting discretionary expenditures (managing accruals) is constrained, and vice versa. Collectively, my findings suggest that overbilling is an important alternative manipulation tool in hospitals.

Acknowledgements

I thank the editor and two anonymous referees for insightful comments that have significantly improved this paper. I also appreciate helpful suggestions from Margaret Abernethy, Ge Bai, Paul Healy, Elina Heese, Thomas Keusch, Frank Moers, Karthik Ramanna, Naomi Soderstrom, Patrick Vorst, and Anne Wyatt. I also thank workshop participants at Monash University, University of Melbourne, University of Queensland, University of Technology at Sydney, University of Western Australia, and the 2013 Global Management Accounting Research Symposium (GMARS) for their helpful comments.

Notes

1 DRGs identify patients with similar conditions who require similar resources for treatment; they are used for reimbursement by major insurance programs such as Medicare (Clemens & Gottlieb, 2015; Heese et al., 2016). A DRG family is a group of DRGs associated with the same underlying ailment. The name of the family (e.g. DRG family 79) reflects the DRG (e.g. DRG 79) that generates the highest reimbursement within the family. Reimbursements for DRGs within a DRG family differ because the treatment of patients with the same underlying ailment (e.g. respiratory ailments) can differ based on the patient’s health status and other complicating factors.

2 For instance, in 2007, the SEC charged Tenet with fraudulent billing and in 2013 it charged the largest hospital in Miami with fraudulent billing (SEC, 2007, 2013).

3 While prior studies (e.g. Zang, 2012) also focus on the presence of high-quality auditors as an AEM constraint, this constraint is very difficult to explore in my setting, as almost all hospitals are audited by Big 4 auditors, and California hospitals are subject to an additional government audit of their financial statements (OSHPD, 2011), making it difficult to disentangle the role of the hospital from that of the government auditor.

4 The Index assigns weights (1, 2, 3, 6) based on comorbidity (e.g. weight 1 for diabetes, weight 6 for metastatic solid tumor) and age (0–5; where 1 point is assigned for each decade starting at 50, e.g. 40–49 = 0; 50–59 = 1). The weighting is based on the number of patients per DRG within each family.

5 Charity care is the difference between established charges for services rendered and the amount paid by or on behalf of the patient, if any. The higher this share, the more costs the hospital faces from treating patients that cannot – or only partly – pay their bills. As an alternative measure, I also use %INDIGENTS_t−1, which is defined as the percentage of indigent patients for whose treatment the hospital received no compensation to the hospital’s total patient population (measured in patient days). The inferences are unaffected (untabulated).

6 EXP is the sum of total research expenditures, total administrative services, total general services, and total education expenditures, and NOCC is the sum of total maintenance expenditures, physicians’ offices and other rentals expenditures, total office building expenditures, child care services expenditures, family housing expenditures, retail operations expenditures, and other non-operating expenditures.

7 where ΔCA is change in current assets; ΔCL is change in current liabilities; ΔCASH is change in cash and cash equivalents; ΔSTD_t is change in debt included in current liabilities; and A_t−1 is total assets in year t−1.

8 Alternatively, I also estimate overbilling, RAM, and AEM using a one-year-ahead prediction model instead of a contemporaneous prediction model. Results from this alternative model are directionally consistent with those using contemporaneous residuals (untabulated).

9 As an alternative estimation technique, I use seemingly unrelated regressions (which takes into consideration the correlation in the error terms across equations), because the earnings management strategies are associated. The inferences are unaffected (untabulated). In addition, recent research by Chen, Hribar, and Melessa (2016) suggests that using a two-step approach to measure overbilling (i.e. I estimate overbilling in the first step and use the residuals from that step as the overbilling measure in the second step) can produce biased results when the regressors from the first step are correlated with regressors in the second step. As a solution, Chen et al. (2016) suggest estimating all coefficients in one step. The results from such single-step model are consistent with those presented in Table 5 (untabulated).

10 Following Heese et al. (2016), I obtain information on FCA settlements from several sources. First, the law firm Fried, Frank, Harris, Shriver & Jacobson, LLP provides a list of settlements for the period 1993–2005 at http://www.friedfrank.com/files/QTam/SettlementsArchive.pdf. Second, the Taxpayers Against Fraud Education Fund lists settlements as of 2004 at http://www.taf.org/resource/fca/statistics. Third, the OIG lists settlements as of 2003 at http://oig.hhs.gov/reports-and-publications/archives/enforcement/false_claims_archive.asp. The Health Care Fraud and Abuse Control Program Reports (available at https://oig.hhs.gov/reports-and-publications/hcfac/index.asp) list FCA settlements as of 1997. In robustness tests (untabulated), I use alternative specifications of PROSECUTION. First, I use only the period following the last violation year until the settlement year as prosecution period and, second, I include all years following the settlement year. The results are robust to these alternative specifications.

11 As the dataset does not include compensation details for individuals, I use the annual sum of salaries, bonuses, and benefits of the top hospital administrators including the Chief Executive Officer, Medical Director, Nursing Director, and their assistants, and scale this sum by the number of full-time equivalent employees (FTEs) in the top administrative team to obtain my measure of compensation (Eldenburg et al., 2011). All variables are panel-specific random parameters.

12 In untabulated robustness tests, I find qualitatively similar results to those reported in Table 5 when I run the tests presented in Table 5 on the subset of hospitals that overbill (as defined by positive residuals). Hospitals that underbill (as defined by negative residuals) use more AEM and RAM.

13 The M-Score is calculated as follows: M-SCORE = −4.84 + 0.920(DSR) + 0.528(GMI) + 0.404(AQI) + 0.892(SGI) + 0.115(DEPI) − 0.172(SGAI) + 4.679(Accruals) − 0.327(LEVI). DSR = (Receivables_t/Net Patient Revenues_t)/(Receivables_t−1/Net Patient Revenues_t−1); GMI = Gross Margin_t−1/Gross Margin_t; AQI = [1 − (PPE_t + Current Assets_t)/Total Assets_t]/[1 − (PPE_t−1 + Current Assets_t−1)/Total Assets_t−1]; SGI = Net Patient Revenues_t/Net Patient Revenues_t−1; DEPI = [Depreciation_t−1/(Depreciation_t−1 + PPE_t−1)]/[Depreciation_t/(Depreciation_t + PPE_t)]; SGAI = (SGA_t/Net Patient Revenues_t)/(SGA_t−1/Net Patient Revenues_t−1); Accruals = (Income before extraordinary items_t − Cash from operations_t)/Total Assets_t; LEVI = Leverage_t/Leverage_t−1.

14 Note that I report results using LOG DRG 79 DISCHARGES in both the AEM and RAM equations. The inferences are unaffected when using LOG DRG 144 DISCHARGES or LOG DRG 296 DISCHARGES instead.