Volume 117, Issue 9 , Pages 629-635, 1 November 2004
Reimbursement denial and reversal by health plans at a university hospital
Article Outline
Purpose
Denial and downgrading of reimbursement for hospital days are two strategies utilized by health plans to maintain profitability. The goal of this study was to describe patterns of discounted reimbursement at a university hospital.
Methods
We performed a retrospective cohort study of consecutive per diem patients hospitalized in 1999. We defined a discounted day as a day fully denied or downgraded and a reversal day as a day reimbursed at a higher level after appeal. The study outcomes included the probability of a discounted day and the probability of a discounted day to be later reversed. Covariance logistic regression was used to compare these outcomes by plan and physician specialty after adjusting for age, sex, race, length of stay, and diagnosis. Correlations with plan characteristics were analyzed.
Results
Of 59,265 hospital days, 6074 days (10.2%) were initially denied or downgraded. On appeal, 1755 discounted days (28.9%) were reversed. The percentage of days discounted per plan ranged from 1.2% to 18.8% (P <0.001), whereas the percentage of discounted days that were later reversed ranged from 23.2% to 85.3% (P <0.001). The qualitative magnitude of these associations and statistical significance were unchanged in adjusted models. Strong correlations were found between the adjusted odds ratio for a discounted day and net profit margin (R = 0.81) and medical loss ratio (R = −0.77).
Conclusion
Denials and downgrades are frequent, with marked variation by health plan. More profitable plans had higher denial and discount rates. Evidence-based standards for denials and downgrades are needed to maintain optimal patient care and the fiscal health of hospitals and health plans.
The growth of managed care and health plans in the United States has greatly increased the financial pressure on hospitals. In response, hospitals have attempted to reduce resource utilization and length of stay through a variety of programs, such as the integration of hospitalists (1, 2). Health plans have implemented a number of utilization management programs to maintain profitability, including advanced approval requirements, concurrent utilization review, and retrospective utilization review (3). Two strategies used to reduce the cost of inpatient hospitalization are complete denial of reimbursement for hospital days and downgrading of reimbursement (4). These reimbursement reductions have been based on a variety of clinical practice guidelines, and are persuasive financial incentives for hospitals to reduce inappropriate hospital admissions and length of stay.
There is limited information describing the magnitude and patterns of denials and downgrades of inpatient reimbursement claims. One study from a community hospital found denial rates of 12% to 19% for inpatient days claimed during a 3-month period in 1996 (5). The goal of our study was to describe the magnitude and variation of inpatient care reimbursement denial, downgrading, and reversal by health plans at a university hospital.
Methods
Study design and sample
We performed a retrospective cohort study of consecutive admissions to the Robert Wood Johnson University Hospital with both admission and discharge between January 1, 1999, and December 31, 1999. The study protocol was approved by the hospital's institutional review board.
We included admissions for which reimbursement was on a per diem basis. This primarily includes patients with managed care or indemnity insurance. We excluded patients with fixed reimbursements per admission, referred to as case-rated reimbursement. Case-rated patients include Medicare and Medicaid patients who are not enrolled in Medicare or Medicaid managed care plans. For these case-rated patients, denial of per diem reimbursement was not possible by definition. We also excluded self-pay and charity care patients and any diagnosis-related groups for which specific health plans had contracted a case-rated reimbursement.
Outcomes
We evaluated hospital denials, downgraded days, and reversal of denials or downgrades according to the following definitions. A denied day was defined as a hospital day for which reimbursement was completely denied. A downgraded day was a hospital day for which reimbursement was downgraded by the plan from the level at which the care was provided, for example, from acute care to skilled nursing. A discounted day was a day that was either fully denied or downgraded. We classified the reason for discounted day based on the primary reason given by the health plan. Only one reason was listed per discounted admission. We grouped reasons for discounting days into the following categories: denial of admission, delayed discharge, delayed service, and other. We defined a reversal day as a previously denied or downgraded day that was reimbursed at a higher level of payment (upgraded) following an appeal. Since the hospital database does not fully characterize reversal decisions to distinguish partial upgrading of reimbursement from full reversals back to the level initially billed, we included an additional outcome—a sustained discounted day—that was defined as a discounted day that remained unchanged after appeal.
Health plan characteristics for the year before our study (1998) were also analyzed. Plan characteristics were unavailable for three of the 13 plans in our study. We analyzed net profit margin, medical loss ratio, and member enrollment. Net profit margin was defined as the ratio of net profit after taxes divided by total revenue. Medical loss ratio was the percentage of premium revenue spent directly on medical services. Member enrollment was the total number of members at the end of 1998.
Data collection
We used the hospital billing database to obtain information on age, sex, race, primary health plan, length of hospital stay, all-patient diagnosis-related groups, and attending physician specialty. Diagnosis-related groups were grouped into 25 major diagnostic categories using the 3M Definitions Manual (6). We used a second database created by the hospital that included the number of days initially discounted, the number of these days reversed, whether the days were fully denied or downgraded or unclassified as to denial/downgrade, and the reason for denial of reimbursement. We used length of hospital stay to define the number of days eligible for per diem reimbursement. In cases where a health plan paid a fixed rate for the first portion of the admission, followed by a per diem rate for subsequent days, only the days eligible for per diem reimbursement were counted.
To avoid identifying health plans, the 13 companies were designated letters ranging from A to M. All primary plans with fewer than 1000 days eligible for per diem reimbursement were grouped together. We matched the health plans with net profit margin, medical loss ratio, and total member enrollment for each plan obtained from Solucient LLC (7).
Statistical analysis
We considered binary outcomes the probability that each visit day was discounted and the probability that a discounted day was later reversed (i.e., either fully reversed or at least upgraded to a higher reimbursement level) on appeal.
Robust covariance logistic regression models were fit using an exchangeable correlation with Proc Genmod to admission days (clustered by patients with exchangeable correlation), to compare the discount and reversal measures. Unadjusted and adjusted models for health plan and physician specialty were fit. Adjusted models included age, sex, race, length of stay, major diagnostic categories, physician specialty, and health plan. There was one major diagnostic category with 18 hospital days and no discounts. Since the logit link modeling requires an outcome, these admissions were removed from the adjusted analyses. Similarly, for the reversals analysis, one major diagnostic category had 21 discounted days and no reversals; these admissions were excluded from the adjusted analysis.
Population discount probabilities were calculated by dividing the total number of discounted days by the total number of hospital days for overall discounts, and the total number of sustained discounts by the total number of hospital days for discounts net of reversals. Population reversal probability was similarly calculated. Confidence intervals for these population probabilities of discount and reversal measures were obtained by back transformation of the confidence interval for the intercept from robust covariance logistic regression models with only an intercept term to hospital days clustered by patient. Nonparametric Spearman correlation coefficients were computed between the adjusted odds ratios of a discounted day and a reversed day, and also between the unadjusted probabilities of discounts and reversals. In addition, Spearman correlation coefficients were calculated for health plan characteristics and the adjusted odds ratios for discounted and reversal reimbursement. All analyses were performed with SAS, version 8.0 (Cary, North Carolina). P values <0.05 were considered significant.
Results
Of the 23,158 total hospital admissions, there were 11,756 (50.7%) per diem admissions among 9604 patients. Most patients (86%) had a single admission, but some had as many as 15 admissions during the study period. Among the 11,756 admissions, 52.6% were men, 71.2% were white, 41.3% were between 41 and 65 years of age, and 15.2% were older than 65 years. Of the 11,756 admissions, 1655 (14.1%; 95% confidence interval [CI]: 13.4% to 14.8%) had at least 1 day initially denied or downgraded. Among these 1655 admissions, 1070 (64.6%; 95% CI: 62.2% to 67.2%) only had days denied, 492 (29.7%; 95% CI: 27.5% to 32.2%) only had days downgraded, and 93 (5.6%; 95% CI: 4.6% to 6.9%) had both days denied and downgraded.
Of 138,710 total hospital days, there were 59,265 days (42.7%) that were per diem, of which 6074 days (10.2%; 95% CI: 9.3% to 11.1%) were initially denied or downgraded. With respect to these 6074 days, 3559 (58.6%; 95% CI: 54.4% to 63.7%) were fully denied (Figure 1). Most reductions in reimbursements were downgrades from acute care rate to a skilled nursing rate. Overall, 28.9% (95% CI: 25.5% to 33.5%) of discounted days were reversed (i.e., fully reversed or partially upgraded). Thus, the rate of sustained discounted days (discounted days net of reversals and upgrades) was reduced from 10.2% to 7.3% (95% CI: 6.4% to 8.0%).
Figure 1.
Flow diagram of denials, downgrades, and reversals of admission claims.
Reasons for denials and downgrades
The most common reasons for denial and downgrading of reimbursement were delayed discharge (64.2%), delay in providing service (16.7%), and inappropriate admission (14.6%). Among delayed discharges, physician-related delays (31.2%) and placement delays (15.2%) were the most common reasons cited by health plans.
Outcomes by health plan
Among plans, the median number of days per admission (median, 3) was consistent, although the mean number of days ranged from 4.3 to 7.4 days. The mean number of days discounted per admission by plan ranged from 0.08 to 1.20 days (P <0.001). The probabilities for an admission day to be discounted varied markedly by health plan (Table 1), with the percentage of days discounted by plan ranging from 1.2% for Plan G to 18.8% for Plan B (P <0.001). When Plan M was designated as the baseline, the odds ratios for the other 12 plans in comparison ranged from 0.54 to 10.58 without adjustment and from 1.01 to 13.67 after adjustment for case mix and other characteristics. Health plan was statistically significant (P <0.001) in the adjusted models, and adjustment did not have a qualitative effect on the odds of denial by most plans.
Table 1. Discounts and Reversals by Health Plan*
| Plan | Discounts | Reversals | Sustained Discounts (Percentage of All Days)*† | ||||
|---|---|---|---|---|---|---|---|
| Percentage of All Days Discounted*† | Odds Ratio (95% Confidence Interval) for a Day to be Discounted | Percentage of Discounted Days Later Reversed†∥ | Odds Ratio (95% Confidence Interval) for a Discounted Day to be Reversed | ||||
| Unadjusted*† | Adjusted†‡§ | Unadjusted†∥ | Adjusted†§¶ | ||||
| A | 17.1 | 9.50 (4.28–21.11) | 13.67(10.22–17.39) | 23.3 | 0.12(0.03–0.52) | 0.12(0.02–0.62) | 13.1 |
| B | 18.8 | 10.58(3.88–28.82) | 10.04(7.17–13.30) | 43.3 | 0.30(0.05–1.88) | 0.43(0.06–3.06) | 10.6 |
| C | 9.8 | 4.97(2.23–11.11) | 6.23(4.66–7.96) | 23.2 | 0.12(0.03–0.52) | 0.13(0.02–0.69) | 7.5 |
| D | 11.7 | 6.09(2.48–14.95) | 8.48(6.19–10.75) | 25.7 | 0.13(0.03–0.72) | 0.14(0.02–0.79) | 8.7 |
| E | 12.7 | 6.62(2.44–17.96) | 11.33(7.98–15.77) | 37.0 | 0.23(0.04–1.27) | 0.13(0.02–0.88) | 8.0 |
| F | 3.8 | 1.80(0.73–4.42) | 2.70(1.96–3.49) | 42.5 | 0.30(0.06–1.50) | 0.27(0.04–1.63) | 2.2 |
| G | 1.2 | 0.54(0.17–1.77) | 1.01(0.64–1.54) | 85.3 | 2.31(0.24–21.80) | 4.03(0.18–89.8) | 0.2 |
| H | 4.1 | 2.10(0.46–9.53) | 3.21(2.05–4.44) | 34.6 | 0.21(0.02–2.58) | 0.31(0.03–3.17) | 2.7 |
| I | 1.8 | 0.95(0.35–2.31) | 1.43(0.81–2.06) | 55.6 | 0.50(0.07–3.31) | 0.45(0.06–3.66) | 0.8 |
| J | 10.0 | 5.09(2.24–11.57) | 6.76(5.01–8.91) | 29.2 | 0.17(0.04–0.77) | 0.13(0.02–0.74) | 7.1 |
| K | 9.3 | 4.93(1.46–16.59) | 9.12(6.29–11.84) | 72.1 | 1.05(0.14–7.91) | 0.89(0.10–8.20) | 2.6 |
| L | 15.2 | 8.26(3.51–19.45) | 12.83(9.24–16.51) | 34.3 | 0.21(0.04–1.03) | 0.20(0.03–1.21) | 10.0 |
| M | 2.3 | Baseline | Baseline | 71.2 | Baseline | Baseline | 0.7 |
* Of 59,265 per diem days. |
† P <0.001 for difference by health plan. |
‡ Of 55,218 per diem days (3947 missing data for adjusted analysis). |
§ Adjusted for attending specialty, length of stay, major diagnostic categories, age, sex, and race. |
¶ Of 5764 downgraded days (310 missing data for adjusted analysis) |
∥ Of 6074 downgraded days. |
The percentage of discounted days later reversed ranged from 23.2% for Plan C to 85.3% for Plan G (P <0.001; Table 1). Using Plan M as the baseline, the unadjusted odds ratio for reversal ranged from 0.12 for Plans A and C to 2.31 for Plan G. After adjustment, the odds ratios ranged from 0.12 for Plan A to 4.03 for Plan G. The association between plan and the odds of reversal remained significant (P <0.001) following adjustment.
There was moderate correlation (R = −0.58, P = 0.04) between discounted days and reversal days for health plans (Figure 2). The correlation between the adjusted odds ratio for a discounted day and reversal day was -0.64 (P = 0.02). The percentage of all days that resulted in sustained discounts (discounts net of reversals) ranged from 0.2% for Plan G to 13.1% for Plan A (Table 1).
Figure 2.
Scatter plot of rates of discounting and reversal by health plan.
Denials and downgrades by specialty
The overall probability for a day to be discounted varied substantially by physician specialty (P <0.001), ranging from 2.6% of admission days for obstetrics and gynecology to 21.2% for family medicine (Table 2). The odds of discounting also varied by specialty when using surgery as the baseline. Some of the odds ratios changed after adjustment for health plan and patient characteristics; in particular, that for pediatrics increased from 0.63 to 1.59 after adjustment.
Table 2. Discounts by Attending Specialty*
| Specialty | Percentage of All Days Discounted*† | Odds Ratio (95% Confidence Interval) for a Day to be Discounted | |
|---|---|---|---|
| Unadjusted†* | Adjusted†‡ | ||
| Family medicine | 21.2 | 2.59(1.75–3.86) | 1.95(1.39–2.74) |
| Medicine | 12.3 | 1.38(1.13–1.70) | 1.25(1.00–1.56) |
| Neurology | 15.6 | 1.84(0.95–3.56) | 1.23(0.70–2.19) |
| Obstetrics/gynecology | 2.6 | 0.26(0.18–0.38) | 0.61(0.27–1.39) |
| Other | 9.9 | 1.08(0.68–1.73) | 1.26(0.79–2.03) |
| Pediatrics | 6.0 | 0.63(0.42–0.96) | 1.59(0.88–2.89) |
| Surgery | 9.4 | Baseline | Baseline |
* Of 59,265 per diem days. |
† P <0.001 for difference by specialty. |
‡ Adjusted for health plan, length of stay, major diagnostic category, age, sex, and race; n = 55,218 per diem days (3947 missing data for adjusted analysis). |
Correlation with health plan characteristics
Net profit margin correlated strongly with the adjusted odds ratio for a discounted day (R = 0.81, P = 0.005), whereas medical loss ratio had a strong negative correlation with adjusted odds ratio for a discounted day (R = −0.77, P = 0.009) by health plan (Table 3). A weaker, nonsignificant correlation of net profit margin (R = −0.34, P = 0.33) and medical loss ratio (R = 0.21, P = 0.57) with adjusted odds ratio were found for a discounted day to be later reversed. However, there was little correlation (R = −0.13) between the adjusted odds of a discounted day and the number of members enrolled in a plan. Correlations between plan characteristics and the probability of discounted days to be later reversed, while never statistically significant, were always in the opposite direction of correlations between the respective characteristic and probability of a discounted day.
Table 3. Correlation of Plan Characteristics and Adjusted Discount Outcomes
| Plan Characteristic | Correlation Coefficient with Adjusted Odds of Outcome for the Plan* (P Value) | |
|---|---|---|
| Day to be Discounted | Discounted Day to be Later Reversed | |
| Net profit margin† | 0.81(0.005) | −0.34(0.33) |
| Medical loss ratio† | −0.77(0.009) | 0.21(0.57) |
| Plan enrollment‡ | −0.13(0.73) | −0.25(0.51) |
† Net profit margin and medical loss ratio data available for 10 of the 13 plans. |
‡ Plan enrollment data available for 9 of the 13 plans. |
Discussion
We found that 10% of inpatient days at a university hospital were denied or downgraded by health plans, most commonly because of delay in discharge. We also found that 29% of initially discounted days were reversed on appeal. The reasons for the high reversal rate are unclear, but suggest that the initial denial process could be improved to reduce these rates. Plans with the lowest rates of discounting had the highest rates of reversals (i.e., correlation between plan discount and reversal rate ∼ −0.6, P <0.05). For example, Plan G had the lowest percentage of discounted days (1.2%) and the lowest adjusted odds of discounting, yet had the highest percentage of discounted days reversed (85.3%) and the highest adjusted odds of reversal. By contrast, plan A, which had the highest adjusted odds of discounted days, also had the lowest adjusted odds of reversal. These patterns suggest that plans with frequent discounting are less likely to reverse their decisions on appeal.
We also examined two measures of financial status of the health plans. The net profit margin, which measures the ratio of net profit to total revenue, had a strong positive correlation with the adjusted odds of a discounted day. The medical loss ratio represents the proportion of premiums attributable to medical care expenses rather than administrative expenses or profit. Lower ratios imply sufficient remaining dollars for administrative expenses and profit. We found that the medical loss ratio had a strong negative correlation with the adjusted odds of a discounted day. These measures of financial health suggest that profitability and discounting of reimbursement are strongly linked.
These findings can be interpreted in two ways. One perspective is that discounting reimbursement via denials and downgrades is an effective strategy employed by health plans to improve short-term profitability. Alternatively, these results may simply indicate that the more profitable health plans are better organized, with the administrative infrastructure to discount reimbursement. Overall, our findings argue against the hypothesis that health plans in greater financial distress are more likely to discount reimbursement; rather, they suggest that profitability is related in part to frequency of discounting.
Appropriate indications for admission to and discharge from the hospital are difficult to define. The ideal criteria would be based on rigorous evidence with clearly defined parameters to identify patients who can be safely treated out of the hospital. Any set of guidelines will be imperfect; some patients will have unnecessary admission and others needing admission will be treated at home. From the perspective of the physician and patient, the tendency will be to err on the side of hospitalization to minimize patient risk and physician liability. From the perspective of the health plan, minimizing hospital days is important to maintain profitability. These circumstances promote differences of opinion unless objective criteria for hospitalization are agreed upon.
Unfortunately, indications for hospital admission and discharge are very difficult to define because they vary by diagnosis and comorbid condition. Data from studies that evaluate large numbers of patients with a particular disorder can be used to create admission and discharge criteria (8, 9, 10, 11). For most diseases, however, there are limited data to develop guidelines, and the process to achieve such evidence-based guidelines is difficult. Thus, many health plans have turned to care guidelines, such as those by Milliman and Robertson to specify the length of stay for a diagnosis (12). However, even these guidelines apply to uncomplicated patients, cannot be applied to all cases, and are not intended as a substitute for medical judgment. Additional research is needed to define the appropriate length of stay and provide criteria for safe discharge. Two studies found that actual practice had substantially higher mean lengths of stay as compared with that recommended by the Milliman and Robertson guidelines for specific diagnoses (13, 14). As pointed out by Bauchner et al (15), many Milliman-Robertson best-practice guidelines are not evidence based. Thus, it is unclear if meeting these benchmarks does ensure patient safety. On the other hand, experience from the past decade when length of stay decreased substantially suggests that there may be safe ways to reduce length of stay.
Given the complexity of hospital care, there are undoubtedly hospital days when tests or procedures are delayed. Inevitably, there are also instances where patients could be treated safely at home or at a lower-cost facility. Innovations, such as case management in the emergency department, clinical pathways, and observation care units, have been developed by some hospitals. In addition, the physician leadership of hospitals can educate and promote guidelines that are truly evidence based. However, the high reversal rates (23% to 85% of all admissions initially discounted) indicate that health plans themselves acknowledged many of their initial denials and downgraded reimbursements that were incorrect.
Our study suggests that the criteria for appropriate duration and acuity of care during hospitalization at our hospital varied by health plan. Although we adjusted for age, sex, race, major diagnostic category, physician specialty, and length of stay, we may not have fully adjusted for differences in disease severity. However, it seems unlikely that this would explain the large differences in discount rates that were found for each of the three outcome measures.
Our study has other limitations. First, some patients had more than one health plan, and we linked discounts to the primary health plan. It is possible that the secondary plan was responsible for the discount, resulting in misclassification bias. However, involvement of multiple plans in reimbursement should diminish (i.e., homogenize) rather than increase observed differences by health plan. Second, practice style may vary within specialties by hospital and state; hence, our results may not be completely generalizable. While the study was conducted at only one hospital, it is unlikely that criteria for appropriate hospitalization would vary by health plan only at this hospital but not vary similarly at other hospitals. Also, determining the frequency of denial because of poor clinical documentation was not possible. Finally, we were unable to characterize fully the extent of the reversals due to limitations in the hospital database. Further investigation of the appeal process and the net outcomes of these appeals are required to determine the net financial effect on hospitals and health plans.
Further studies are needed to develop criteria for discharge and evaluate the probability of untoward clinical outcomes for the most common diseases in every specialty. Pending adequate data, criteria for hospitalization should be developed jointly by clinicians and health plans to ensure that patients are not discharged prematurely and that the fiscal health of hospitals and health plans are protected.
References
- . The emerging role of “hospitalists” in the American health care system . N Engl J Med . 1996;335:514–517
- Reorganizing an academic medical service: impact on cost, quality, patient satisfaction, and education . JAMA . 1998;279:1560–1565
- . Back to the drawing board: new directions in health plans' care management strategies . Health Aff (Millwood) . 2002;21:210–217
- . Reducing insurer-denied days of care . Healthc Financ Manage . 1996;50:52; 54,56-58
- . Reducing the number of days for which insurers deny payment to the hospital: one primary objective for a newly configured department of case management . Soc Work Health Care . 1998;28:31–49
- 3M. All Patients Diagnosis Related Groups Definitions Manual. 8th ed. 3M Health Information Systems 3M Health Information Systems, Wallingford, Connecticut;; 1990.
- HCIA. The Guide to the Managed Care Industry, 2000 Edition. Baltimore, Maryland: HCIA, L.L.C.; 1999.
- A prediction rule to identify low-risk patients with community-acquired pneumonia . N Engl J Med . 1997;336:243–250
- The hospital discharge decision for patients with community-acquired pneumonia (Results from the Pneumonia Patient Outcomes Research Team cohort study) . Arch Intern Med . 1997;157:47–56
- The hospital admission decision for patients with community-acquired pneumonia (Results from the pneumonia Patient Outcomes Research Team cohort study) . Arch Intern Med . 1997;157:36–44
- Time to clinical stability in patients hospitalized with community-acquired pneumonia: implications for practice guidelines . JAMA . 1998;279:1452–1457
- . Just how short can hospital stays be . Med Econ . 1994;71:37–38
- Pediatric Milliman and Robertson length-of-stay criteria: are they realistic . Pediatrics . 2000;105:733–737
- . An analysis of 25 Milliman & Robertson guidelines for surgery: data-driven versus consensus-derived clinical practice guidelines . Ann Surg . 1998;228:579–587
- . Milliman and Robertson—going in the wrong direction . Pediatrics . 2000;105:858–859
PII: S0002-9343(04)00482-6
doi:10.1016/j.amjmed.2004.06.025
© 2004 Elsevier Inc. All rights reserved.
Volume 117, Issue 9 , Pages 629-635, 1 November 2004
