International Journal for Quality in Health Care 15:139-146 (2003)
© 2003 International Society for Quality in Health Care
Paper |
Hospital experience and outcomes for esophageal variceal bleeding
Divisions of 1General Internal Medicine
2Gastroenterology, Department of Medicine, Johns Hopkins School of Medicine, Baltimore, MD
3Department of Health Policy and Management, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA
Objective. Although higher hospital volume has been associated with better outcomes for many surgical procedures, this relationship does not appear to hold for most common medical diagnoses. We evaluated whether there is a volume–outcome relationship for a rarer and higher-mortality medical diagnosis, esophageal variceal bleeding.
Design. Cross-sectional retrospective study of hospital discharge data.
Setting. All Maryland hospitals from 1992 through 1996.
Study participants. All patients with diagnosis codes for both esophageal variceal bleeding and cirrhosis in relevant diagnosis-related groups.
Main outcome measure. Mortality for esophageal variceal bleeding. We classified hospitals by tertiles of admissions as high (>17 cases of variceal bleeding per year), medium (12–17 cases per year) or low (<12 cases per year) volume.
Results. There were seven high-volume, 13 medium-volume, and 29 low-volume hospitals. Overall in-hospital mortality was 15%. After multiple regression was used to adjust for differences in age, sex, ethnicity, emergency room admission, use of procedures, complexity, Medicaid status, transfer status, and clinical variables associated with mortality from variceal bleeding, there were no significant differences between the high-, medium-, and low-volume hospital groups in in-hospital mortality (16%, 15%, and 13%, respectively). There were also no significant differences in hospital charges ($17 000, $15 000, and $14 000, respectively) or length of stay (8.5, 8.7, and 7.8 days, respectively) between hospital volume groups.
Conclusions. The volume–outcome relationship may not pertain to some medical diseases such as esophageal variceal bleeding. Alternatively, the biases inherent in research using administrative data may make this relationship appear weaker for some medical than for surgical diagnoses in this type of study.
Keywords: administrative data, cirrhosis, end-stage liver disease, esophageal variceal bleeding, hospital volume, mortality, tertiary care
Hospital volume is a frequently studied potential predictor of quality of care. Although it may be correlated with physician volume or teaching status, it may also capture an institution’s multidisciplinary expertise and availability of supportive facilities. In addition, new advances in therapy may be adopted more quickly at hospitals that treat higher volumes of certain illnesses. The volume–outcome relationship is of particular importance because regionalizing care for high-risk surgeries or diagnoses might reduce population mortality [1–4]. Higher volume might also improve the efficiency of care and reduce resource use.
A recent literature review found that, for conditions with a study of sufficient quality to be included, the best-quality study showed that high-volume centers had lower mortality for 13 procedures but for only one medical diagnosis [1]. Although higher volume hospitals might admit patients with more severe illness, there have been no published studies showing worse outcomes with higher volume [3]. In addition, studies of teaching hospitals have also shown significantly lower mortality for surgeries but very small or insignificant benefits for common medical illnesses [5–7], despite increased resource use [8].
Esophageal variceal bleeding is a common complication of cirrhosis with a mortality of 17–42% per bleeding episode [9,10]. Preventing death from variceal bleeding requires adequate and timely resuscitative measures, specific treatments aimed at arresting the bleeding quickly, and skilled management of the liver disease and secondary complications [9]. In Maryland, the presence of trained specialists such as hepatologists is correlated with hospital volume. We therefore applied the methods used to study hospital volume for surgeries to esophageal variceal bleeding because, compared with broader medical diagnoses that have shown small or insignificant differences by volume [3], variceal bleeding is rarer, mortality is higher, and care is more dependent on procedures and multidisciplinary expertise that may not be available in smaller hospitals.
Specifically, we performed an analysis of Maryland hospital discharge data to assess whether increased hospital experience with variceal bleeding was associated with decreased in-hospital mortality, length of stay, or hospital charges for such patients.
Methods
Study design and data sources
This study was a retrospective, cross-sectional analysis of administrative claims data from 1992 to 1996 using the Maryland Health Services Cost Review Commission (MHSCRC) database. This database includes information on all admissions to acute-care non-federal Maryland hospitals, including International Classification of Diseases: Ninth Revision, Clinical Modification (ICD-9-CM) [11] codes for up to 15 discharge diagnoses and procedures, diagnosis-related group (DRG) codes, patient demographics (age, sex, ethnicity, and insurance status), in-hospital mortality, total hospital charges, and hospital length of stay. The database does not include other clinical, endoscopy, physical exam or laboratory information, or data on the patient’s course before or after the hospitalization.
To assess the quality of the data, the validity of our inclusion criteria and the presence of potential confounders not captured in the claims data, we also used the Johns Hopkins Hospital case-mix database. This includes the same information as the MHSCRC database, as well as unique patient identifiers allowing linkage to discharge summaries. The Johns Hopkins Joint Committee on Clinical Investigation approved this study.
Definition of cases of esophageal variceal bleeding
We developed a clinical definition of esophageal variceal bleeding based on the literature and consultation with experts. Based on this, we then developed an algorithm for identifying patients who fit this clinical definition as accurately as possible using the codes available in the administrative database. Our clinical definition of esophageal variceal bleeding was endoscopically verified identification of actively bleeding varices in a patient with cirrhosis, significant enough to be the primary reason for non-elective admission to a hospital. Figure 1 shows the algorithms that we used to best identify patients who fit this clinical definition using codes available in the database. Because there may be errors in coding medical diagnoses in administrative data, we performed our primary analyses with a specific (conservative) coding case definition. We performed a sensitivity analysis to these assumptions by repeating the analysis with a sensitive (less conservative) coding case definition.
|
We excluded liver transplant patients, because they would have very different mortality and resource use. We also excluded elective admissions in all analyses. For the specific coding case definition, we also excluded cases with a primary diagnosis or procedure code inconsistent with the diagnosis of variceal bleeding and any cases that did not have a procedure code for esophagoscopy. Because we were unable to determine the ultimate outcome of patients who were transferred out to other hospitals, these patients were also excluded.
Validation of the case definitions
To estimate the specificity and sensitivity of our primary coding case definition, we reviewed all discharge summaries from the Johns Hopkins Hospital discharge database with a code for variceal bleeding in the defined time period. We then determined whether each discharge summary fit our definition of variceal bleeding. To estimate the sensitivity of the coding case definition, we reviewed a random sample of 20 cases of each of five diagnosis and procedure codes that might identify patients with variceal bleeding but did not include a code for it. We also reviewed records for the accuracy of the MHSCRC’s data on in-hospital mortality and looked for potential factors in the process of care that might explain differences in mortality.
Data analysis
Descriptive analyses were performed for pertinent independent variables including age, sex, race, Medicaid status (a proxy for lower socioeconomic status), All Patient Refined (APR)-DRG [12] complexity score, transfer status, cause of liver disease, type of procedures performed, comorbidities, known predictors of mortality in variceal bleeding, and hospital volume. Descriptive analyses were also performed for the three outcome variables (in-hospital mortality, hospital length of stay, and hospital charges). Total charges were adjusted for inflation to 1996 dollars using the medical care component of the consumer price index [13].
The distribution of hospital volume was not continuous and there were no clear divisions by exploratory data analysis. Therefore, we divided hospitals into tertiles based on the distribution of admissions, consistent with the quality recommendations in a recent review article on the hospital volume–outcome relationship [1]. The reason for this recommendation is to reduce bias from variability between individual hospitals. To compare the characteristics of patients between hospital volume groups, we performed bivariate analyses using the chi-square test and ANOVA. To assess how the outcomes related to the independent variables, we performed bivariate analyses using the chi-square test when mortality was the dependent variable, and the student’s t-test when log-transformed charges or length of stay was the dependent variable. Because both length of stay and total charges were highly skewed, we used the modified Park test [14] to determine the best regression method, and used the robust generalized linear model with the gamma family and log link to compare these variables [15].
We then performed multiple logistic and linear regression with the robust generalized linear model with the gamma family and log link to determine whether the outcomes were related to hospital volume. We accounted for patient clustering within hospitals using multilevel modeling and robust variances [16]. We adjusted for age, sex, Caucasian ethnicity, Medicaid status as a proxy for lower socioeconomic status, transfer from another hospital, and year of admission. We adjusted for complexity of illness using the most widely used system for determining risk of mortality, the APR-DRG system [12]. We also adjusted for markers of the severity of liver disease that have been associated with increased mortality in variceal bleeding, based on a review of the literature. We defined these markers using ICD-9-CM codes, as shown in Table 1. These included the presence of alcoholic liver disease, hepatocellular carcinoma, hepatorenal syndrome, encephalopathy, ascites, and coagulopathy [17–22]. We also adjusted for whether or not a transjugular intrahepatic portosystemic shunt (TIPS) was performed; since this may have been in the causal pathway, we conducted a sensitivity analysis without adjusting for TIPS.
|
We performed several additional sensitivity analyses. We repeated the analyses using the Charlson–Deyo score [22] instead of the APR-DRG complexity score for risk adjustment. We repeated the analysis by categorizing hospital volume differently, by bifurcating by number of admissions. Finally, since the ultimate outcome of patients transferred out to other hospitals could not be obtained from the database, we also reanalyzed the data by including these patients with randomly imputed mortality rates, total charges, and length of stay using the distribution of values for patients transferred in to other hospitals.
Analyses were conducted using STATA Version 6.0 software (Stata Corporation, College Station, TX).
Results
Validation of the main case definition, mortality, and potential explanations for differences in mortality
Of 154 cases satisfying the inclusion criteria for the main (specific) case definition, 76% of discharge summaries were definitely consistent and 10% were possibly consistent with the definition of variceal bleeding. Of a sample of 100 cases not identified by the most sensitive case definition, but with an ICD-9-CM code for a related diagnosis or procedure, only 7% had a discharge summary consistent with variceal bleeding.
The mortality from the discharge summaries consistent with the specific diagnosis of variceal bleeding was 21%, similar to the 19% unadjusted mortality determined for this hospital using the MHSCRC database. When we reviewed summaries for potential processes of care that may have affected mortality, we also found that many of the patients who died had care stopped or withdrawn during the last days of life because they were so ill that further treatment was not felt to be beneficial.
Patient characteristics
Of the 2883 hospitalizations with codes for variceal bleeding, 1654 fit the main case definition (58%) (Figure 1); this group of patients is described below. Forty-nine hospitals had at least one admission that met the inclusion criteria for the specific case definition; seven were high volume, 13 were medium volume, and 29 were low volume. Two hospitals had no admissions that met the inclusion criteria. The distribution of patients by hospital is shown in Figure 2. The average age of patients was 56 years (range 18–93 years), 64% were male, 72% were Caucasian, and 19% had Medicaid insurance.
|
Fifty-eight percent had a code for cirrhosis from alcohol, 4% for hepatorenal syndrome, 16% for hepatic encephalopathy, 30% for ascites, and 6% for TIPS. The distribution of patient characteristics by hospital group is shown in Table 1. Significant differences for the highest volume hospitals included a lower percentage of Caucasian patients, younger patients, more patients with Medicaid insurance, more patients with the highest complexity score, and more patients transferred in from other hospitals.
Overall mortality was 15%; the range among hospitals is shown in Figure 3. Among all hospitals, overall median hospital length of stay was 6 days (range among hospitals 1–10.5 days) and overall median total hospital charges (adjusted to 1996 dollars) were $9000 (range among hospitals $2000–20 000).
|
Changes over time
There was no significant change in the annual number of cases statewide over the 5-year period, nor in the percentage of cases in the high-volume group. There was a significantly increased percentage of patients transferred between hospitals (from 4 to 8%, P = 0.02) and a trend toward a higher percentage receiving TIPS (from 4 to 8%, P = 0.10). Most patient characteristics were stable, except for increases in the percentages of patients with hepatorenal syndrome and ascites. In-hospital mortality and inflation-adjusted hospital charges did not change over the 5-year period, but length of stay did decrease (from 9.8 to 7.5 days, P < 0.001).
Role of TIPS
Surgical portacaval shunts are now rarely performed in the state of Maryland; almost all patients who need a shunt now receive TIPS performed by interventional radiologists (personal communication). Most TIPS were performed at a few medium- or high-volume hospitals; 22% of hospitals performed no TIPS, and the highest percentage of patients receiving TIPS at a hospital was 50%. Patients who received TIPS had significantly higher mortality (22% versus 14%, P = 0.02), inflation-adjusted total hospital charges ($42 000 versus $13 000, P < 0.001), and length of stay (16 versus 8 days, P < 0.001) than patients who did not receive TIPS. Patients with TIPS also had a higher complexity score (64% versus 34% with a complexity score of 4, P < 0.001), and a higher proportion had some markers of severity of illness, such as ascites (45% versus 29%, P = 0.001). The differences in total charges and length of stay persisted both for patients who died and those who survived hospitalization.
Relation of hospital volume to outcomes
In bivariate analyses, as summarized in Table 2, male sex, non-Caucasian ethnicity, emergency room admission, higher complexity score, and TIPS were all significantly associated with higher in-hospital mortality for esophageal variceal bleeding. As summarized in Table 3, many clinical factors were also associated with mortality, length of stay, and hospital charges.
|
|
Neither in-hospital mortality nor total hospital charges were significantly different by hospital volume group after adjusting for differences in patient characteristics, as shown in Figure 4 (P = 0.89). Male gender, age >65 years, emergency room admission, year of admission, a high complexity score, hepatorenal syndrome, and encephalopathy were independent predictors of mortality in the multivariable model. After adjusting for the independent variables listed above, there was also no significant difference in length of stay by hospital volume group. The risk adjustment in the model showed good calibration (Hosmer–Lemeshow, P = 0.57) and discrimination (c-statistic = 0.82) [24].
|
Sensitivity analysis
There was no significant difference in mortality by hospital volume in any of the alternative analyses that we performed to test the robustness of our model. Our results did not change when we used a more sensitive case definition. Using hospital volume as a continuous variable, including patients who had been transferred, and using the Charlson-Deyo comorbidity score did not significantly change the results of the regression analyses either.
Discussion
Many studies using administrative data and similar methods have found that higher-volume hospitals have better outcomes for varied surgical procedures. We therefore hypothesized that after adjusting for confounders and severity of illness, higher hospital volume would be associated with better outcomes for patients with esophageal variceal bleeding, a high-mortality medical illness with a high incidence of procedure use. There are several potential explanations for why we did not find a hospital volume–outcome relationship. First, this relationship may not exist for all procedures and settings. A recent systematic review found only 13 procedures where this relationship held [1], and another large study found relationships for only five of 16 procedures evaluated [25]. High-risk, non-acute procedures may be more likely to show a benefit [1].
Secondly, we found that risk adjustment accounted for much of the difference between hospital volume groups, and many other studies of hospital volume have performed limited risk adjustment. A recent large study of eight intermediate-complexity surgical procedures within the Veterans Administration did not find any effect of hospital volume either, possibly because they were able to perform more accurate risk adjustment with their prospectively collected data than in most other studies [26].
Our results are consistent with other recent studies that have shown greater benefits for surgical than medical procedures in comparisons of teaching and non-teaching hospitals [6,8]. Although our sample size was comparable to many of the surgical studies, we did not have sufficient power to detect the difference of <5% which was found for many of these diagnoses; differences for procedures are often much higher. Although the reasons for the hospital volume–outcome relationship and differences between conditions have not been well explored, it is possible that high-volume hospitals are better at coordinating care and taking advantage of the available expertise in some clinical situations [3].
It is also possible that the potentially inadequate risk adjustment inherent in any retrospective study favors procedures over diagnoses. This may be particularly true for most of the studies of the volume–outcome relationship, which use administrative data. Administrative data is limited in its ability to account for the severity of the primary illness requiring the surgical procedure or medical admission. A major difference between surgical and medical diagnoses is the issue of in-patient transfers versus outpatient referrals. Although transferred patients are often more ill and more costly than non-transferred patients [27], one study showed that referred patients (those from outside the local catchment area for a hospital) had 33% lower mortality, possibly because patients of higher socioeconomic status and lower severity of illness may be more likely to travel for surgery [28]. We did adjust for transfers in our study, but very few comparisons of procedures have adjusted for referrals or distance traveled. Finally, anecdotal evidence suggests that some lower-volume hospitals may have better mortality because they systematically identify and quickly transfer patients more likely to benefit from the care available at the higher-volume centers.
Mortality is a limited indicator of the spectrum of quality of care, and has a different meaning in relatively healthy patients undergoing elective surgery and in patients with a life-threatening complication of a serious illness. In fact, the conditions that have shown the strongest relation between volume and in-hospital mortality are those where the long-term prognosis is most limited (surgery for pancreatic and esophageal cancer) [1]. One recommendation of the recent Institute of Medicine report on this topic [3] is that studies investigate processes of care to determine why the hospital volume–outcome relationship does or does not exist for a particular condition. Our discharge summary review revealed that many patients were so ill that further care was felt to be futile and was stopped or withdrawn.
Our study had several limitations that are common to most studies evaluating the volume–outcome relationship. Although our database contained many of the predictors that have been shown to be associated with mortality from variceal bleeding, these may be undercoded because the frequency is somewhat lower than in clinical studies [17,19]. However, many factors in our model, such as the complexity score, were highly predictive of mortality so we should have been able to account for much of the severity of the variceal bleeding episode.
We also found in our validation that, although our case definition appeared to capture most of the cases of variceal bleeding in the database, some cases were inaccurately coded as patients with variceal bleeding. The proportion of cases that were inaccurately coded is consistent with the findings of a recent systematic review that rarer diagnoses tend to have higher rates of inaccuracy [29]. Ethnicity and Medicaid insurance are also an incomplete representation of socioeconomic status, which may be an important predictor of mortality for a disease like cirrhosis which disproportionately affects underserved populations. Finally, although we could not account for any post-discharge effects of differences in care between hospital groups, comparisons of hospitals do not tend to differ when comparing in-hospital and 30-day mortality [30].
Although many studies have shown a volume–outcome relationship for surgical procedures, most of this research has used administrative data with limited risk adjustment. It is unclear why this relationship should only hold for some procedures, and why the relationship does not exist for most medical diagnoses, such as esophageal variceal bleeding. We found that biases that differ between medical and surgical procedures and information, such as withdrawal of treatment, that are not available in administrative databases may be important reasons for these differing findings, and deserve further investigation.
Address reprint requests to Sydney Morss Dy, Room 609, 624 N. Broadway, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD 21205, USA. E-mail: sdy{at}jhsph.edu 
Accepted for publication November 27, 2002.
References
- Dudley RA, Johansen KL, Brand R et al. Selective referral to high-volume hospitals: estimating potentially avoidable deaths. JAMA 2000; 283: 1159–1166.
[Abstract/Free Full Text] - Birkmeyer JD, Lucas FL, Wennberg DE. Potential benefits of regionalizing major surgery in Medicare patients. Eff Clin Pract 1999; 2: 277–283.[Medline]
- Hewitt M, for the Committee on Quality of Health Care in America and the National Cancer Policy Board. Interpreting the volume–outcome relationship in the context of health care quality: Workshop summary. Washington, DC: Institute of Medicine, 2000.
- Begg, CB, Cramer, LD, Hoskins WJ, Brennan MF. Impact of hospital volume on operative mortality for major cancer surgery. JAMA 1998; 280: 1747–1751.
[Abstract/Free Full Text] - Rosenthal G, Harper D, Quinn D et al. Severity-adjusted mortality and length of stay in teaching and nonteaching hospitals: results of a regional study. JAMA 1997; 278: 485–449.[Abstract]
- Taylor D, Whellan D, Sloan F. Effects of admission to a teaching hospital on the cost and quality of care for Medicare beneficiaries. N Engl J Med 1999; 340: 293–299.
[Abstract/Free Full Text] - Cooper G, Chak A, Harper D et al. Care of patients with upper gastrointestinal hemorrhage in academic medical centers: A community-based comparison. Gastroenterology 1996; 111: 385–390.[CrossRef][ISI][Medline]
- Yuan Z, Cooper GS, Einstadter D, Cebul RD, Rimm AA. The association between hospital type and mortality and length of stay: a study of 16.9 million hospitalized Medicare beneficiaries. Med Care 2000; 38: 231–245.[CrossRef][ISI][Medline]
- Graham D, Smith JL. The course of patients after variceal hemorrhage. Gastroenterology 1981; 80: 800–809.[ISI][Medline]
- Jenkins SA, Shields R, Davies M et al. A multicentre randomised trial comparing octreotide and injection sclerotherapy in the management and outcome of acute variceal haemorrhage. Gut 1997; 41: 526–533.
[Abstract/Free Full Text] - US Department of Health and Human Services. ICD-9-CM. International Classification of Diseases, 9th revision; Clinical Modification, Fifth Edition. Los Angeles, CA: Practice Management Information Corporation, 1997.
- 3M Health Information Systems. 3M All Patient Refined DRG Software. http://www.3m.com/market/healthcare/his/us/products/apr_drg/brochure.html (accessed September 5, 2002).
- Department of Commerce. Indexes of medical care prices: 1970 to 1996. In: Office GP, ed. Statistical abstract of the United States. Washington, DC: Department of Commerce, 1996.
- Park R. Estimation with heteroscedastic error terms. Econometrica 1966; 34: 888.[CrossRef]
- Newson R. sg114: rglm—Robust variance est. for generalized linear models. In: Newton H, ed. The Stata Technical Bulletin Reprints. College Station, TX: Stata Corporation, 1999, pp. 181–190.
- Paterson L, Goldstein H. New statistical methods for analysing social structures: An introduction to multilevel models. Br Educ Res J 1991; 17: 387–393.
- Garden O, Motyl H, Gilmour W et al. Prediction of outcome following acute variceal hemorrhage. Br J Surg 1985; 72: 91–95.[Medline]
- Faivre J, Milan C, Joyeux J et al. Factors of early prognosis in medically treated haemorrhages of upper gastrointestinal tract in cirrhotic patients. Statistical study. Path Biol 1981; 29: 399–404.
- Le Moine O, Adler M, Bourgeois N et al. Factors related to early mortality in cirrhotic patients bleeding from varices and treated by urgent sclerotherapy. Gut 1992; 33: 1381–1385.
[Abstract/Free Full Text] - Spence R, Terblanche J, Kahn D et al. Ascites and serum bilirubin value predict survival in acute variceal bleeding. S African Med J 1990; 77: 334–338.
- Gatta A, Merkel C, Amodio P et al. Development and validation of a prognostic index predicting death after upper gastrointestinal bleeding in patients with cirrhosis: A multicenter study. Am J Gastroenterol 1994; 89: 1528–1536.[Medline]
- Chalasani N, Kahi C, Francois F et al. Model for end-stage liver disease (MELD) for predicting mortality in patients with acute variceal bleeding. Hepatology 2002; 35: 1282–1284.[CrossRef][ISI][Medline]
- Charlson M, Pompei P, Ales K et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chron Dis 1987; 40: 373–383.[CrossRef][ISI][Medline]
- Iezzoni L, ed. Risk Adjustment for Measuring Healthcare Outcomes. Chicago, IL: Health Administration Press, 1997.
- Hannan E, O’Donnell JF, Kilburn H Jr et al. Investigation of the relationship between volume and mortality for surgical procedures performed in New York State hospitals. JAMA 1989; 262: 503–510.[Abstract]
- Khuri SF, Daley J, Henderson W et al. Relation of surgical volume to outcome in eight common operations: results from the VA National Surgical Quality Improvement Program. Ann Surg 1999; 230: 414–429.[CrossRef][ISI][Medline]
- Clough JD, Kay R, Gombeski WR Jr et al. Mortality of patients transferred to a tertiary care hospital. Cleve Clin J Med 1993; 60: 449–454.[ISI][Medline]
- Ballard DJ, Bryant SC, O’Brien PC et al. Referral selection bias in the Medicare hospital mortality prediction model: are centers of referral for Medicare beneficiaries necessarily centers of excellence? Health Serv Res 1994; 28: 771–784.[ISI][Medline]
- Campbell SE, Campbell MK, Grimshaw JM, Walker AE. A systematic review of discharge coding accuracy. J Public Health Med 2001; 23: 205–211.
[Abstract/Free Full Text] - Rosenthal GE, Baker DW, Norris DG et al. Relationships between in-hospital and 30-day standardized hospital mortality: implications for profiling hospitals. Health Serv Res 2000; 34: 1449–1468.[ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||