We examined clinical outcomes and estimated health care costs for 490 seriously ill patients who developed renal failure requiring dialysis. We stratified patients according to an objective estimate of prognosis and determined the cost-effectiveness of initiating dialysis and continuing aggressive care in patients who were at low, average, and high risk for dying within 6 months.

Methods

Top Methods Results Discussion Author & Article Info References

Patients

We studied seriously ill patients enrolled in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT) who had renal failure after enrollment and were treated with hemodialysis or peritoneal dialysis. Patients who were already undergoing dialysis at the time of hospital admission were excluded.

The Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment was a prospective study of outcomes, preferences, and decision making for seriously ill hospitalized adults. Full descriptions of the objectives and methods of SUPPORT have been published elsewhere [6, 7]. Inpatients were enrolled prospectively from June 1989 to January 1994 at five geographically diverse academic medical centers.

During the intervention phase of SUPPORT, clinicians who had been randomly assigned to the intervention group were given information about their patients' prognoses and preferences for care; skilled nurses facilitated communication between patients and clinicians. Because the intervention did not affect mortality or resource use, we included patients in the intervention and control groups in our analyses. Patients were eligible for SUPPORT if they were 18 years of age or older and met defined criteria for at least one of nine diagnostic categories: acute respiratory failure, chronic obstructive pulmonary disease, congestive heart failure, cirrhosis, non-traumatic coma, metastatic colon cancer, advanced non-small-cell lung cancer, multiorgan system failure with sepsis, or multiorgan system failure with malignancy. Specific diagnostic criteria [6] were designed to identify patients at late or advanced stages of illness who, on average, had an estimated 6-month survival probability of approximately 50%. All patients were screened for eligibility at hospital admission; patients in intensive care units were screened daily. Eligible patients who were discharged or died within 48 hours of study entry were excluded from SUPPORT and were not available for inclusion in our analyses.

The design of SUPPORT was approved by the institutional review boards of the participating hospitals, and informed consent was obtained verbally before interviews were conducted with patients, their families, and their physicians.

Data Collection

Data were collected by chart abstraction and interview. Research nurses reviewed medical records throughout patient hospitalization and abstracted data on patient diagnoses; comorbid conditions; and Acute Physiology Scores [8] on study days 1, 3, 7, 14, and 25. Comorbid conditions (such as diabetes mellitus, congestive heart failure, and stroke) were abstracted by chart reviewers by using a list that had been developed as part of the APACHE (Acute Physiology and Chronic Health Evaluation) II scoring system; a comorbidity score was calculated by a simple count of comorbid conditions [9]. The Acute Physiology Score has been previously shown to predict in-hospital mortality; a higher score indicates increased risk.

Detailed data on resource utilization that included a revised version of the Therapeutic Intervention Scoring System (TISS) were also collected on study days 1, 3, 7, 14, and 25. The Therapeutic Intervention Scoring System, an additive measure of intensity of resources used, assigns 1 point for minor interventions (such as pulse oximetry, physical therapy of the chest, and peripheral intravenous therapy) and 2 to 4 points for more substantial interventions (such as intubation, thrombolytic therapy, endoscopy, and surgery). This system has been shown to be a valid and reliable measure of hospital resource use [10, 11]. Hospital charts were also reviewed periodically during patient hospitalizations and retrospectively to determine whether hemodialysis or peritoneal dialysis had been initiated.

Trained interviewers questioned patients and their surrogates. A surrogate was defined as the person who would make decisions on the patient's behalf if the patient were unable to do so. Patients were not interviewed if they were unable to communicate because of coma, intubation, cognitive impairment, or other reasons. Data gathered at the initial interview, conducted with the patient or surrogate between study day 2 and day 6, provided information on the patient's functional status 2 weeks before study entry, as measured by a modified version of the Katz Index of Activities of Daily Living [12, 13]. Interviewers also obtained this information 6 months after study entry. When a patient's report was unavailable, we substituted a surrogate's report of the patient's abilities to perform activities of daily living. Surrogate reports of patient functional status before admission were used for 279 of the 345 patients for whom we gathered information at baseline and for 32 of the 94 patients for whom we reported functional status at the 6-month follow-up visit.

Interviewers asked patients to rate their quality of life as excellent, very good, good, fair, or poor. To measure utilities (quality-of-life weights) [14-16], patients were asked to state the amount of time spent in excellent health that they would equate with living for 12 months in their current state of health (a time-tradeoff question). A patient who was willing to trade 12 months in his or her current health for 6 months in perfect health, for example, would have a utility of 0.5 (6 ÷ 12).

Cost Estimates

For each patient, total hospital charges were gathered from the billing systems of participating hospitals for the index hospitalization and for all readmissions to the same hospital during the 6 months after study entry. Hospital costs were estimated by adjusting charges using Medicare cost-to-charge ratios for UB82 (Uniform Bill 1982) cost centers at each participating hospital. For the four patients for whom no billing information on the index hospitalization was available, we imputed hospital costs on the basis of a linear regression model that considered length of stay and average TISS score after dialysis was initiated. We transformed all estimated costs into 1994 U.S. dollars using the medical component of the consumer price index.

From the cost of the index hospitalization, we estimated hospital costs incurred after dialysis was initiated. Hospital costs have been shown to be highly correlated with the product of the length of stay and average TISS score [7]. We assumed that this relation would hold for the portion of the hospitalization that occurred after the initiation of dialysis. Therefore, we estimated the cost of the portion of the index admission incurred after the initiation of dialysis by using the following equation:

Cost = total cost of index hospitalization x [length of stay x TISS score after dialysis] ÷ [length of stay x TISS score for entire hospitalization]

Data on hospital costs were collected for readmissions to the study hospitals during the 6-month SUPPORT period and were used to estimate additional hospital costs incurred during the first 6 months of follow-up. These cost estimates were based on the mean number of readmissions between discharge from the hospital and the end of the 6-month study period multiplied by the mean cost of hospitalization for patients who had only one readmission (62% of patients had no readmissions, 22% had one, 13% had two to three, and 3% had four to five). We estimated the cost of each readmission on the basis of patients who had only one readmission, because available cost data included only aggregated costs for all readmissions.

Hospital costs beyond 6 months were estimated from Medicare data that reported annual hospital costs for patients with end-stage renal disease who were treated with dialysis [17]. We assumed that although most patients in our cohort would not require long-term dialysis, the nature of the illnesses that qualified them for SUPPORT suggested that they have chronic illness and would require hospitalization at a rate similar to that of patients who had a serious illness, such as dialysis-dependent kidney disease. We also assumed that patients would not receive a kidney transplant.

For survivors who remained dialysis-dependent, we estimated outpatient costs of long-term dialysis as follows. In a comprehensive cost analysis comparing the costs of dialysis with the costs of transplantation, we estimated the total annual cost of dialysis to be $32 800 in 1989 dollars ($46 322 in 1994 dollars) [17]. Because 41% of those costs were for inpatient care, we subtracted 41% from this Figure to avoid double counting inpatient hospital costs. Therefore, we assumed that the annual outpatient cost would be $27 330 for survivors who required long-term dialysis. A recent study of patients in the intensive care unit who developed renal failure requiring dialysis found that after 1 year, 33% of survivors remained dialysis-dependent [5]. On the basis of these data, we estimated that for our cohort overall, long-term dialysis would cost $9019 (0.33 x $27 330) per year (Appendix Table).

Life Expectancy Estimates

We estimated life expectancy by extrapolating from available survival data. Data on survival beyond the 6-month SUPPORT study were updated by using the National Death Index; patients were followed for as long as 4.4 years. On the basis of observed survival beyond 6 months, we estimated life expectancy beyond 6 months using an exponential function [18, 19].

Statistical Analysis

Descriptive Statistics and Prognostic Model

To describe patients at baseline and at the end of 6 months, we tabulated distributions for categorical variables and the median and 25th and 75th percentiles for continuous variables. For the 4301 patients enrolled in the observation phase of SUPPORT, a Cox proportional-hazards model of time to death yielded estimates of 6-month mortality for each patient [20]. The SUPPORT model estimates were based on age, diagnosis, comorbid illness, and 11 physiologic variables. To improve calibration of these estimates for our subgroup of 490 patients, we developed a second Cox proportional-hazards model in which the dependent variable was survival and the independent variable was log (–log) of the estimates of 6-month survival from the SUPPORT model. From this second model, an estimate of the probability of surviving 6 months after treatment with dialysis was calculated for each patient. Patients were stratified according to their prognostic estimates, and actual 6-month survival for each stratum was calculated.

Cost-Effectiveness Analysis

Our analyses considered only direct medical costs; we did not estimate additional societal costs or savings. We included estimated costs of inpatient hospitalization and long-term outpatient dialysis in our analyses. Inpatient hospital costs did not include physician fees. The decision to withhold dialysis in the setting of serious illness with renal failure requiring dialysis was assumed to be part of a decision to withdraw care and allow death to occur. Therefore, cost and survival after withholding dialysis were assumed to be zero because the costs of caring for patients after withholding dialysis and discontinuing life-prolonging treatments are negligible compared with the costs of continuing aggressive care. Data from SUPPORT show that 98% of patients in whom dialysis was withheld died within 6 days of hospital discharge and 100% died within 77 days of discharge. We expect that if the decision to withhold dialysis were made in the context of a general decision to shift the focus of care from life extension to comfort, survival would be even shorter and the costs of caring for such patients until death would be very low. As a result, the cost per year of life gained after initiating dialysis would be equivalent to the incremental cost-effectiveness of choosing dialysis and aggressive treatment as opposed to withholding dialysis and allowing death to occur.

To calculate the cost of initiating dialysis and continuing aggressive care, we calculated the cost of the portion of the index hospitalization that occurred after the initiation of dialysis, as described above. To this sum, we added the projected costs of future inpatient hospital care and long-term dialysis for each year of survival (Appendix Table). All costs and years of survival incurred after year 1 were discounted at a rate of 3%. To estimate the incremental cost per year of life saved, we divided the total discounted costs by the total discounted years of survival. We performed this process for subgroups of patients stratified according to their prognostic categories. Cost-effectiveness for the entire cohort was estimated by compiling estimates for each subgroup. To adjust years of life for quality and to allow for estimations of costs per quality-adjusted life-year gained, we multiplied the denominator by 0.84. This value is the mean utility (quality-of-life weight) as ascertained from patients at 6-month interviews. Incremental cost-effectiveness ratios were rounded to the nearest $100 per quality-adjusted life-year.

In a separate analysis, we used utility values from interviews conducted on study days 3, 7, 14, 25, 60, and 180 to estimate each patient's quality-adjusted survival during the first 6 months of follow-up. For this secondary analysis only, we imputed missing utility values from regression models with survival time as the independent variable and calculated quality-adjusted survival time through 6 months for each patient on the basis of the area under the utility-time curve. Because the results were similar to those obtained when only the mean 6-month utility value was used, we report the results of the simpler approach.

Sensitivity Analyses

Sensitivity analyses were performed to determine how different assumptions would affect our results. We varied each of our cost estimates from 50% to 200% of our baseline estimates. The cost of the index hospitalization was the largest component of total costs, and we had primary index hospitalization cost data for each patient. We explored further how sensitive our results were to variation in estimates of the index hospitalization cost by calculating cost-effectiveness ratios on the basis of the upper and lower limits of the 95% CIs for costs of index hospitalization in each risk group. We varied patient utilities from the 10th to the 90th percentile for patients interviewed (0.5 to 1.0) and varied the discount rate from 0% to 10%. Finally, because two studies [21, 22] of renal failure in critically ill patients found that as few as 12% of patients required long-term dialysis, we repeated our analysis using this estimate rather than our baseline estimate of 33%.

Results

Top Methods Results Discussion Author & Article Info References

Patient Characteristics

Of the 9105 seriously ill patients enrolled in SUPPORT, 490 patients who were not already undergoing long-term dialysis had dialysis initiated during the study period. The median patient age was 61 years, 58% of patients were female, and 69% of patients had acute respiratory failure or multiorgan system failure with sepsis as the qualifying diagnosis for SUPPORT (Table 1).

Clinical Outcomes

Median survival from the time dialysis was initiated was 32 days; only 27% of patients were alive 6 months after dialysis was started (Table 2). Patients who survived had a median of one dependency in activities of daily living, and 62% of survivors rated their quality of life as "good" or better (Table 2).

For the five prognostic groups defined by survival estimates from the recalibrated SUPPORT prognostic model, actual survival closely approximated predicted survival (Table 3). Compared with the remaining 396 patients, the 94 patients in the best prognostic category (41% to 60% probability of surviving 6 months) were younger (53 compared with 59 years of age) and had lower APACHE III Acute Physiology Scores at day 3 (52 compared with 68). Compared with all higher-risk patients, patients in the best prognostic group were more likely to have SUPPORT diagnoses of congestive heart failure (16% compared with 4%) or acute respiratory failure or multiorgan system failure with sepsis (79% compared with 67%) but were less likely to have multiorgan system failure with malignancy (1% compared with 16%) or cirrhosis (0% compared with 8%) (P < 0.01 for all comparisons). The median number of comorbid illnesses did not differ for the 94 patients with the best prognoses (P = 0.17).

Cost-Effectiveness Analysis

Baseline Analyses

The estimated median costs (in 1994 dollars) of the index hospitalization from the time dialysis was initiated until death or discharge was $31 991. Costs of future hospital care beyond the 6-month follow-up period were estimated to be $17 466 per year (Table 2). Although median costs are reported, mean costs were used to calculate cost-effectiveness ratios. Overall, the incremental cost per quality-adjusted life-year saved for initiating dialysis and continuing aggressive care in seriously ill hospitalized adults was $128 200 (Table 3). The cost per quality-adjusted life-year varied widely across prognostic strata. For patients whose probability of surviving 6 months was 10% or less, the incremental cost of initiating dialysis and continuing aggressive care was $274 100 per quality-adjusted life-year. For patients with the best prognoses (a 40% to 60% probability of surviving 6 months), the incremental cost was $61 900 per quality-adjusted life-year.

Sensitivity Analyses

Sensitivity analyses revealed that varying our assumptions widely resulted in only small changes in estimated cost per quality-adjusted life-year. Our results were most sensitive to large changes in the estimate of the cost of the index hospitalization. When we decreased the index hospitalization cost estimates by 50%, the overall cost per quality-adjusted life-year was $80 000; when we increased our estimates by 100%, the cost per quality-adjusted life-year was $224 700 (Table 4). In the second sensitivity analysis of index hospital costs, which incorporated the upper and lower bounds of the 95% CIs for the index hospitalization costs, the ranges of the cost-effectiveness ratios for each risk group were as follows: $53 100 to $70 600 per quality-adjusted life-year for patients with a 41% to 60% probability of 6-month survival; $61 200 to $92 800 per quality-adjusted life-year for patients with a 31% to 40% probability of 6-month survival; $70 400 to $101 300 per quality-adjusted life-year for patients with a 21% to 30% probability of 6-month survival; $99 900 to $145 300 per quality-adjusted life-year for patients with an 11% to 20% probability of survival; and $200 800 to $347 400 per quality-adjusted life-year for patients with a 0% to 10% probability of survival. When we did sensitivity analyses using extreme assumptions, we found that simultaneously halving all cost estimates and assuming a higher mean utility of 1.0 reduced the overall cost per year of life saved to $53 900 (range across prognostic categories, $26 000 to $115 100 per year of life saved) and that doubling all cost estimates and assuming a lower mean utility of 0.5 increased the overall cost per quality-adjusted life-year to $430 800 (range across prognostic categories, $207 900 to $920 900 per quality-adjusted life-year).

Discussion

Top Methods Results Discussion Author & Article Info References

We studied the costs and outcomes of 490 seriously ill hospitalized patients who underwent dialysis for renal failure. Median survival was 32 days, and only 27% of patients were alive 6 months after dialysis was initiated. For the few patients who survived for 6 months, quality of life and functional status were fairly good.

Overall, treatment with dialysis and continued aggressive care in this setting was expensive, with an estimated cost of more than $128 000 (1994 dollars) per quality-adjusted life-year saved. For the large group of patients (21%) with the worst prognostic estimates (10% probability of surviving for 6 months), the estimated incremental cost of initiating dialysis and continuing aggressive care was $274 100 per quality-adjusted life-year, more than twice the overall cost estimate. However, for the subgroup with the best prognoses (a 41% to 60% probability of surviving for 6 months), this treatment strategy is more cost-effective, at $61 900 per quality-adjusted life-year. Our results did not change substantially even with large changes in baseline cost estimates.

Although analyses of Medicare expenditures [17] have shown that dialysis is, on average, a relatively cost-effective medical intervention at $46 322 (1994 dollars) per year of life saved, the cost-effectiveness of initiating dialysis and continuing aggressive care when renal failure develops as a complication of serious illness has not been previously studied. The appropriateness of routinely initiating dialysis in this setting has been questioned, and efforts have been made to prospectively identify patients who are unlikely to survive despite aggressive treatment [2, 4, 5, 23, 24]. Patients who require dialysis in the setting of hospitalization for serious illness are in a unique position: Their short-term prognosis is poor, but without dialysis and continued intensive care death will almost certainly occur. For these patients, the cost-effectiveness of starting dialysis and continuing aggressive care is driven not primarily by the cost of dialysis but by the high cost of continuing intensive care in the face of a low probability of survival.

Our findings show that compared with other medical interventions, dialysis is not cost-effective for patients with average prognoses who develop renal failure in the setting of serious illness: For patients with worse-than-average prognoses, dialysis and continuing care cost several times more than the commonly cited upper limit for cost-effective care, $50 000 per quality-adjusted life-year [25]. By comparison, treatment with tissue plasminogen activator instead of streptokinase for coronary thrombolysis in myocardial infarction costs about $34 200 per year of life saved (1994 dollars) [26], coronary artery bypass surgery rather than medical therapy for left main coronary artery disease costs $8300 per quality-adjusted life-year (1994 dollars) [27, 28], and medical therapy for severe hypertension costs $23 800 per quality-adjusted life-year (1994 dollars) [26, 29].

Our study has several limitations. First, our costs represent estimated rather than actual costs and are not based strictly on hospital resource use. However, previous analyses of SUPPORT data demonstrated that the estimates of hospital costs that we used correlate closely with the product of length of hospital stay and TISS score, resource-based measures of health care utilization [7]. For patients who survived the index hospitalization, we projected future hospital costs from cost data available from the 6-month follow-up period and the literature, but sensitivity analyses showed that our results are not sensitive to wide variation in estimates of downstream costs. Our cost estimates did not include physician fees, outpatient costs, or costs of readmissions during the initial 6 months of follow-up to nonstudy hospitals; it is therefore likely that our results represent underestimates of true medical costs. To adjust years of life saved for their quality, we assumed that patients who were not interviewed had values similar to those who completed the 6-month interview. If patients who could not be interviewed had lower utilities and if dialysis costs were underestimated by not including physician fees or outpatient costs, the cost-effectiveness of dialysis and continued aggressive care would be even less favorable. The estimates of survival, quality of life, and initial hospital costs that we used in our cost-effectiveness analysis were obtained by using data from five geographically diverse medical centers. Although we acknowledge that clinical outcomes and health care costs for patients treated at other centers may be different, we believe that our data are the best that are currently available with which to perform a cost-effectiveness analysis.

An additional limitation was that SUPPORT included only seriously ill patients who had any of nine selected diagnoses; patients with AIDS were excluded. Therefore, our results may not be generalizable to patients with other serious illnesses. Finally, we stratified patients according to a recalibrated version of the SUPPORT model. Because the recalibration of the SUPPORT model does not alter its discrimination, the model can be expected to perform as well in an independent sample at separating low-risk from higher-risk patients. However, we would not expect our estimates to be as well calibrated in another group of patients.

Our findings provide clinicians and patients and their families with important information about outcomes for seriously ill patients who have renal failure requiring dialysis. For the few patients who survive this complication, functional outcomes and quality of life are fairly good. Except for the subgroup of patients with the best prognoses, however, dialysis and continued aggressive care in this setting are not consistent with many experts' standard for cost-effective medical care. Our findings highlight an increasingly common ethical dilemma: When a health care technology has the potential to save life, but at very high cost, should patients continue to have unlimited access to such expensive care or should care be rationed (either implicitly or explicitly) so that resources can be diverted to patients who are likely to derive more benefit from them? And if such medical care is to be rationed, should the rationing be done by physicians, hospitals, insurance companies, the government, or some combination of these decision makers? Societal debate on these important issues is needed, and information on the cost-effectiveness of specific treatments should inform such discussions.

Presented in part at the National Meeting of the Society of General Internal Medicine, Washington, D.C., May 1996.

Dr. Desbiens: Chattanooga Unit, University of Tennessee College of Medicine, 921 East Third Street, Suite 400, Chattanooga, TN 37403.

Dr. Connors: Department of Health Evaluation Sciences, University of Virginia School of Medicine, Box 600, Charlottesville, VA 22908.

Drs. Teno and Lynn: The Center to Improve Care of the Dying, 1001 22nd Street, NW, Suite 700, Washington, DC 20037.

Dr. Wenger: University of California, Los Angeles, Department of Medicine, B-564/Factor, 10833 Le Conte Avenue, Los Angeles, CA 90095-1736.

Dr. Wu: Johns Hopkins University, Health Services Research Center, 624 North Broadway, Baltimore, MD 21205.

Dr. Fulkerson: Duke University Medical School, Box 3121, Durham, NC 27710.

Dr. Tsevat: Section of Outcomes Research, University of Cincinnati Medical Center, 231 Bethesda Avenue, Cincinnati, OH 45267-0535.