Hospital Costs and Severity of Illness in Three Types of Elective Surgery

This study was approved with exemption from informed consent by Stanford University's Human Subjects Committee. We chose three common surgical procedures that require a wide range of hospital resources, have patients with various levels of illness severity, and are performed at both tertiary and nontertiary hospitals. We used International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) procedure numbers 81.54 (total knee replacement), 51.23 (laparoscopic cholecystectomy), and 45.73–45.8 (colectomy) to obtain a list of all patients who had undergone one of these procedures at Stanford University Medical Center from September 1993 to September 1994. Only patients admitted to the hospital the day of or the day before surgery were included. For total knee replacement, only unilateral, primary operations were included.

For purposes of calculating our preliminary sample size, we assumed that we were comparing two groups-ASA PS 1 and PS 2 patients with ASA PS 3 patients. Using a random-number table, we selected 60% of patients having one of the three operations. This resulted in (1) enough patients to detect a 20% difference (power, 0.8; significance level, 5%; two tailed) in hospital costs between two groups of patients for the procedure with the smallest sample size (colectomy, n = 30), and (2) greater power for the other two surgical procedures with more patients.

We retrospectively reviewed the charts of 224 patients. For each patient, basic demographic data including age, sex, and insurance type were recorded from the hospital administration database. For the analyses, Medicare and Medicaid patients were grouped together. We included insurance type in our model to control for referral and payer patterns. The ASA PS scores were abstracted from the anesthesia record. The anesthesiologists assigning the ASA PS were not aware of the study. We matched the medical record data with hospital administrative data to confirm that the operation being studied was the principal surgical procedure of the patient's hospitalization. Emergency cases were excluded.

The comorbidity index developed by Charlson was designed and validated as a prognostic taxonomy for comorbid conditions that, singly or in combination, alter medical outcomes or resource use. [17]This comorbidity index weighs both the number and seriousness of comorbid diseases (appendix 1 Table 6). We choose this index because:(1) unlike the ASA PS, the Charlson index is derived from administrative data; and (2) it is widely used in health services research. [18,19]We used ICD-9-CM codes from Stanford University Medical Center's administrative database to compute a comorbidity index score for each patient. [18] 

Cost data were obtained from a widely used hospital cost accounting software developed by Transition Systems Inc. (TSI; Boston, MA). We examined actual hospital costs instead of patient charges because charges are an inaccurate measure of hospital resource use. [20]We used total hospital costs instead of variable costs because, at our institution, total hospital costs have a strongly positive linear relation with variable costs (unpublished data). The TSI software combines clinical and financial data to determine total patient costs. Costs for all resources expended during a patient's hospitalization are summed to determine the patient's total hospital costs. Equipment and supply costs, support services (i.e., medical records), and labor effort are entered into TSI by hospital department managers. Transition Systems Inc. cost standards are continuously tested for accuracy and recalibrated based on actual revenues and expenses. Physician professional fees were not included in our analysis because they were not available to us.

Resource consumption profiles were established for each procedure. For each patient, we determined total hospitalization costs, operating room costs, anesthesia costs, and length of hospital stay. Total hospital costs were the sum of costs attributed to the operating room, anesthesia, postanesthesia care unit, patient ward, laboratory, pharmacy, surgery admission unit, intensive care unit, blood blank, and radiology department.

Operating room costs reflected facility costs associated with the operating room time necessary to complete each case and costs accrued in producing specific patient services. Operating room costs were computed as the sum of (1) a basic rate calculated according to how long the patient is in the operating room (as recorded by operating room nurses), which includes labor (two clinical staff), supplies (i.e., laparotomy sponges), and other (i.e., linen) common to all procedures, and (2) incremental resources (i.e., knee prosthesis) specific to the surgical procedure.

Anesthetics were delivered by staff anesthesiologists and anesthesia housestaff. Anesthesia costs for each patient included all intraoperative drug costs, airway supplies, intravenous and blood administration supplies, invasive pressure monitoring, regional anesthesia supplies, and salaries of anesthesia technicians. Fixed costs (e.g., depreciation of the anesthesia machine) were also included when calculating total costs.

We asked the following questions. What is the relation between ASA PS and the Charlson comorbidity index? We used the Spearman rank correlation coefficient to evaluate this relation.

Does either measure of severity of illness predict costs of elective surgeries? For each surgical procedure, we used backward-elimination multiple regression to build models for predicting (1) total hospital costs, (2) operating room costs, (3) anesthesia costs, and (4) length of stay. Explanatory candidate variables included patient age (y), sex, ASA PS 1–3, Charlson comorbidity index, and type of insurance (Medicare/Medicaid, managed care, or indemnity). In backward-elimination multiple regression, the model begins with all of these independent variables. Then the variable that causes the smallest reduction in R²is removed. Next the remaining variables are tested to determine if additional variables can be removed without significantly increasing the residual variance. At each step, the variable that produces the smallest increase in residual variance is removed, then the process is repeated, until there are no variables in the equation that could be removed without significantly increasing the residual variance. We repeated these analyses on the pooled data of all 224 patients.

Data evaluation was performed using Statistical Analysis Software (SAS Institute, Cary, NC). Statistical significance was indicated by P < 0.05. Results are presented as mean (95% confidence interval) unless otherwise indicated.

(Table 1) shows the characteristics of the patients studied and the distribution of patients by procedure, ASA PS, and Charlson comorbidity index score. There were only three ASA PS 4 patients, one for each surgical procedure, whose data we combined with the ASA 3 patients. This decision did not change the results.

We found a statistically significant weak positive correlation (R = 0.34) between ASA PS and Charlson comorbidity index (P < 0.001). All ASA PS 1 patients had a Charlson comorbidity index of 0 or 1 and a mean Charlson score of 0.05 (95% CI +/- 0.06). The mean Charlson comorbidity index for ASA PS 2 patients was 0.40 (95% CI +/- 0.25) and 1.47 (95% CI +/- 0.91) for ASA PS 3 patients.

We found no consistent relation between hospital costs and either of the two severity-of-illness indices. Type of insurance was not a predictor of costs.

Neither the ASA PS nor the comorbidity index developed by Charlson predicted total hospital costs for laparoscopic cholecystectomy (unadjusted mean =$3,778; 95% CI +/- 299) or colectomy (unadjusted mean =$13,614; 95% CI +/- 3,019). The Charlson comorbidity index, but not ASA PS, predicted hospital costs for knee replacement (unadjusted mean =$18,788; 95% CI +/- 573). For knee replacements, a unit increase in the Charlson comorbidity index score resulted in an estimated increase in hospital costs of $1,229 (95% CI +/- 788; P = 0.003).

When we pooled all the data, only the type of surgical procedure performed was an independent predictor of total hospital costs (Table 2).

The unadjusted mean operating room costs equaled $1,640 (95% CI +/- 118), $3,316 (95% CI +/- 512), and $8,616 (95% CI +/- 296), respectively, for laparoscopic cholecystectomy, colectomy, and knee replacement. The ASA PS, but not the Charlson comorbidity index, predicted operating room costs only for colectomies:$666 (95% CI +/- 551) per unit increase in physical status score (P = 0.026;Table 3). The ASA PS and Charlson comorbidity index did not predict operating room costs for patients having knee replacement or laparoscopic cholecystectomy.

The mean unadjusted anesthesia costs were $412 (95% CI +/- 34) for knee replacement, $464 (95% CI +/- 47) for colectomy, and $306 (95% CI +/- 10) for laparoscopic cholecystectomy. The only positive relation we found was between ASA PS and anesthesia costs for colectomies (Figure 1). Anesthesia costs for colectomies increased $216 (95% CI +/- 118) per unit increase in ASA PS (P = 0.001;Table 3). In this operation, longer case times, increased monitoring, and more frequent use of combined general and regional anesthesia accounted for the increase in intraoperative anesthesia costs (Table 4).

The ASA PS was not predictive of increased length of hospital stay for any of the three surgical procedures. Table 5summarizes length of stay data (unadjusted to explanatory variables) by surgical procedure and ASA PS. The Charlson comorbidity index was an independent predictor of length of hospital stay only for patients having knee replacements. Length of stay increased 0.62 (95% CI, 0.40) days per unit increase in Charlson comorbidity index score (P =.005). Hospital deaths did not contribute to shorter lengths of stay.

The ASA PS classification is nonlinear in that the increment from ASA PS 1 to ASA PS 2 probably is not equal to the increment from ASA PS 2 to ASA PS 3. Because of this, we pooled ASA 1 and 2 patients and compared that cohort to ASA 3 patients. This decision was not helpful in predicting costs.

We did not analyze costs by surgeon (n = 12) or anesthesiologist (n = 19) because of the small number of cases per provider.

To manage inpatient health-care costs appropriately, hospital managers and physicians need to know what factors influence perioperative costs for elective surgeries. Providers often claim that they generate greater hospital costs because their patients are more seriously ill. Patients who are more severely ill may experience more complications (thereby increasing costs) or the illnesses causing the greater severity of coexisting disease may worsen as a result of the surgical procedure and hospitalization. Because the incidence of complications is low in patients having elective surgery, patients in our study who were more ill could have greater total hospital costs due to more preoperative interventions (i.e., laboratory tests), larger intraoperative costs due to longer case times, more extensive monitoring, and greater drug costs, or from longer and more intensive postoperative surveillance.

If severity of illness did affect the cost of elective surgeries, this factor should be used when adjusting reimbursement so health systems and physicians do not resist caring for more seriously ill patients. If severity of illness is not considered, providers who select, or who tend to be selected by, more seriously ill patients may be penalized. We found that two different severity-of-illness measures, ASA PS scores 1–3 and the comorbidity index developed by Charlson, did not reliably predict costs or length of stay for three types of low-risk, elective inpatient surgical procedures.

Although statistically significant, the positive correlation between ASA PS and Charlson comorbidity index scores was low (r = 0.34). ASA PS 3 patients had more variation in Charlson comorbidity index scores than did ASA PS 1 patients. This may reflect that (1) the ASA classification is a global measure of severity of illness and functional status, whereas the Charlson index reflects the number of comorbid conditions and not the degree to which each condition affects the patient's function; and (2) defining a patient with severe systemic disease that limits activity (but is not incapacitating) is more difficult than defining a healthy patient. In the Charlson index, certain comorbid variables, such as metastatic cancer, are weighted heavily, which, depending on the surgical procedure, may or may not increase hospital costs. Other important conditions (i.e., congestive heart failure) may have a greater effect for some surgeries than suggested by their weights in the Charlson index. This index was useful in predicting costs and length of stay only for patients having knee replacement. Investigators are studying whether any individual variable (or the total number of diagnoses) in the Charlson index is a significant predictor of length of stay. [21] 

Imperfect severity-of-illness adjustment is a possible explanation for our findings. We used two distinct and separately derived indices intended to measure severity of coexisting disease, not the severity of disease of the target organ for which the patient had surgery. By studying two measures we determined how well ASA PS performs compared with another severity-of-illness measure in predicting hospital costs. In addition to the ASA PS and the comorbidity index developed by Charlson, other severity-of-illness measures include the Acute Physiology and Chronic Health, [22]Medical Illness Severity Grouping System (MedisGroups), [23]and Computerized Severity indices. [24]These mathematically derived, frequently proprietary systems are available to hospitals, payers, and governments to produce illness severity-adjusted patient outcomes. These severity measures use detailed clinical data from patients' medical records or data from a hospital's discharge records to predict resource consumption or inpatient death.

Our findings that severity of illness is an inconsistent predictor of elective surgical costs correspond with those of other studies. Thomas and Ashcraft [1]compared six commercial severity models for their ability to predict costs among hospitalized surgical and medical patients. The fraction of total variance in costs (R squared) explained by the severity models was low, ranging from 0.06 for the Acute Physiology and Chronic Health II index to 0.18 for the Computerized Severity Index. We found isolated relations between severity of illness and anesthesia and operating room costs. For example, ASA PS and age explained 49% of the variance in operating room costs for patients having colectomies. The increased operating room costs for patients with greater ASA PS reflect longer case times for these patients, because case time is a key input when calculating operating room costs.

That ASA PS predicts costs for some, but not all, elective procedures corresponds with findings of previous investigations. Cullen [14]found that ASA PS was not associated with increased hospital stay in patients having total hip replacement. However, there was a positive association between length of stay and ASA PS for patients having open cholecystectomy or transurethral prostatectomies. The relation between severity of illness and surgical costs may vary by the risk associated with the procedure, by the population of patients having that procedure, and by the choice of severity-of-illness measure. Thus, researchers working with administrative databases need to use multivariate analyses to derive comorbidity weights tailored to the patient population, condition, and outcome being studied.

There is no reliable standard that determines in which ASA PS group a particular patient belongs. Owens and associates [25]studied how 255 board-certified anesthesiologists classified 10 hypothetical cases. They found that only 6 of the 10 cases were rated consistently. The ASA PS is used as a billing modifier. This may inject a bias into the score assignment. It is possible that subjectivity in ASA PS assignment makes it an unreliable severity-of-illness measure.

The subjectivity of raters is also an important methodologic issue for other severity-of-illness measures. However, the comorbidity index defined by Charlson uses objective clinical criteria to assign scores. Charlson took detailed clinical data from patient records to mathematically derive and validate the comorbity index. This index and others such as Acute Physiology and Chronic Health, MedisGroups, and the Computerized Severity Index have been shown to predict health outcomes. [17,22–24]The fact that these indices can predict outcome suggests that subjectivity alone cannot explain the inability of these indices to predict costs.

The procedures we studied ranged from a relatively low-cost operation (laparoscopic cholecystectomy) to a higher cost surgery (knee replacement). We used hospital costs instead of charges, because charges do not accurately reflect hospital costs. [20]An inadequate cost accounting system could mask any existing relation between severity of illness and hospital costs. This is unlikely, given the sophistication, testing, and acceptance of the system used in this study.

We studied noncritically ill patients having elective surgery. Our results differ from those of previous studies of critically ill patients. Those studies suggested that a small number of very ill patients consume a disproportionately large share of resources. [26,27]For example, patients who required prolonged intensive unit care after coronary artery bypass graft had greater severity-of-illness scores as measured by the Acute Physiology and Chronic Health II index. [26]Oye reported that 8% of the most severely ill intensive care unit patients consumed most of the hospital's critical care resources. [27] 

The greatest increases in costs may occur in patients who suffer adverse outcomes. However, the number of patients in this study was not large enough to detect differences in adverse outcomes and subsequent costs because the quality of care delivered was high or the patients' risks were low. We could not determine if elective surgical costs increase for ASA PS 4 patients, because fewer than 2% of the patients we studied were classified as ASA PS 4. However, ASA PS 4 patients comprise such a small fraction (< 3%) of persons having elective procedures that this category is unlikely to be a useful predictor. [11] 

For elective surgery, resources are expended primarily in accomplishing the surgical procedure and managing its consequences, rather than managing the patient's coexisting or coincidental diseases. Perhaps when surgeons report that their patients are more seriously ill, they are referring to surgical (not coexisting) disease that is not captured by common severity-of-illness measures.

Variability in resources required for surgical care (i.e., operating room costs) and length of stay may predict variability in total hospital costs. Wennberg and others [28,29]have documented large variations in medical and surgical practice patterns that cannot be explained by patients' diseases. At the time of the study, patient care protocols were in effect for postoperative nursing issues. The emergence at our institution of perioperative clinical pathways (standardized diagnostic and treatment protocols addressing physician-directed interventions) after completion of the present study highlights the importance of variability in care as a primary determinant of costs.

A cost analysis of total knee replacement showed that operating room costs represented 40% of the average $18,788 for the procedure. The acquisition cost of the prosthesis comprised two thirds of the operating room costs. Other studies have confirmed that the prosthesis accounts for approximately 25% of the total cost of total knee replacement. [30]The large cost of the prosthesis (and variability in prosthesis choice) may obscure any costs related to managing underlying or associated medical illnesses.

Various anesthesia techniques can produce similar outcomes with different costs. [31,32]Our study suggests that for some procedures (i.e., laparoscopic cholecystectomy), incremental anesthesia costs may be related more to the practitioner's choice of anesthetic than to the patient's severity of illness. On the other hand, increasing levels of coexisting illness for patients having colectomy resulted in longer anesthesia and surgery time and more frequent invasive monitoring and use of epidural anesthesia. Knowing the ASA PS alone explained 33% of the variability in anesthesia costs for these patients. Even if anesthesia costs can predict hospital costs, the overall magnitude of the effect is small, because intraoperative anesthesia costs are less than 6% of total hospital costs. [20]Dexter and Tinker [33]presented evidence that changes in anesthetic care are unlikely to reduce total hospital costs for cases similar to the ones we studied.

Our study was conducted at a 660-bed, university-affiliated, tertiary care medical center with 19,000 surgical cases each year. The results reflect the medical practices of the providers studied and may not be applicable to other settings. We did not control for the potentially confounding effects of individual physician practices on hospital costs, because of the many providers studied and the relatively few cases per physician. This study did not assess provider fees. We also did not include very sick patients because we only studied patients classified as ASA PS 1–3.

Because the taxonomy used to describe costs affects the results of economic evaluations of health care interventions, [34–36]a different method to measure and quantify costs may have yielded different results. Another possible confounder is that patients with a higher severity-of-illness score were cared for in lower-cost units. Hospital cost systems are based on cost appraisals of various department managers who follow specified cost accounting guidelines. However, whether allocated costs are equivalent for the same item used in different surgical units deserves further study.

To determine when physical status can be used to predict hospital costs, other frequent, or expensive, surgical procedures should be analyzed to determine their cost structure and the cause of cost variances. The postulated confounding effect of variability in care should be confirmed. We are quantifying this effect by investigating the relation between severity of illness and costs with and without standardized perioperative protocols. More detailed studies are required to identify other determinants of cost variation, such as nonmedical (disposition) problems or the severity of surgical disease requiring surgery. The effects of anesthetic management on subsequent events that might affect hospital costs also merit further study.

Surgical costs arise in a complex manner, and their prediction is difficult. Patient-related factors have only a negligible effect on hospital costs. Other factors such as variability in resource use (i.e., the cost of the knee prosthesis as selected by the surgeon) may be more important than severity of illness in producing costs. Our study of patients undergoing three types of low risk, inpatient surgical procedures suggests that, at least in these groups, severity of illness, as categorized by ASA PS (1 to 3) or by the Charlson comorbidity index, does not predict surgical costs or length of hospital stay. These relations may be different in medical or critically ill patients. These data suggest that hospital reimbursement for low-risk elective surgery need not be based on patient acuity. Reimbursement by procedure alone, independent of severity of coexisting disease, may work reasonably well because resources are expended primarily to manage the consequences of surgical decisions and techniques, rather than to manage the patient's related or coexisting diseases.

The authors thank Sandra Dewey for help with data collection and manipulation.