Two-Way Referral Bias: Evidence from a Clinical Audit of Lymphoma in a Teaching Hospital

J Clin Epidemiol Vol. 51, No. 2, pp. 93–98, 1998 Copyright  1998 Elsevier Science Inc. All rights reserved.

0895-4356/98/$19.00 PII S0895-4356(97)00244-8

Two-Way Referral Bias: Evidence from a Clinical Audit of Lymphoma in a Teaching Hospital O. Paltiel,* I. Ronen, A. Polliack, and L. Epstein Department of Social Medicine and School of Public Health, Hadassah/Hebrew University and Department of Hematology, Hadassah Medical Center, Jerusalem, Israel ABSTRACT. The objective of this study was to evaluate the effect of referral bias in a clinical audit of lymphoma in a university hospital. We compared demographic and clinical characteristics as well as survival for Jerusalem residents (local) and referred (distant) patients diagnosed from 1987 to 1992 and treated in our institution. Referred patients were younger (p , 0.0001), and less likely to be immigrants (p , 0.0001), than local patients. Aggressive non-Hodgkin’s lymphomas (NHL) were more common in the referred population ( p 5 0.015). Survival for Hodgkin’s disease was consistently better for local patients, but for patients with NHL the findings were reversed. In this study referred patients differed in their clinical and sociodemographic characteristics but did not consistently exhibit a worse outcome than that of local patients. The unpredictable nature of referral bias may be due to better functional status or resources among referred patients, or to selective referral for procedures such as bone marrow transplantation. While reports on the natural history of disease from tertiary institutions may be biased by referral patterns, the direction of the bias is not uniform. j clin epidemiol 51;2:93–98, 1998.  1998 Elsevier Science Inc. KEY WORDS. Lymphoma, epidemiology, clinical audit, referral bias

INTRODUCTION ‘‘Referral filter bias,’’ as defined by Sackett [1], occurs when the concentration of rare causes, multiple diagnoses, and ‘‘hopeless’’ cases increases as a group of ill patients are referred from primary to secondary to tertiary care. Published reports from tertiary institutions may thus reflect the consequences of referral rather than an accurate depiction of the distribution of disease or its outcome. Increased disease complexity or severity is often associated with worse outcomes [2] although this has not been reported consistently [3]. We performed a clinical audit on the outcome of lymphoma patients diagnosed and/or treated in our institution and noted that nearly half were from outside of the Jerusalem area. In the Israeli health care system, with its universal coverage under the Health Insurance Law, patients would, as a matter of course, be directed to an institution close to their place of residence by their local Sick Fund. Outside of the Jerusalem area the choice of tertiary institutions, including teaching hospitals, is wide, the closest teaching hospital being approximately 45 kilometers from the city. The decision to send a patient to Hadassah, therefore, would reasonably constitute referral, whether by *

Address for correspondence: Dr. Ora Paltiel, MDCM, MSc., FRCPC, Dept. of Social Medicine, Hadassah Medical Organization, POB 12000, Jerusalem, Israel, 91120. Accepted for publication on 13 October 1997.

the physician or the patient him/herself. This report compares demographic and disease characteristics as well as survival between ‘‘local’’patients and those from outside the Jerusalem district catchment area (hereafter considered ‘‘referred’’) in an attempt to examine whether significant referral bias occurred. SUBJECTS AND METHODS Hadassah University Hospital is a 915-bed teaching hospital situated on two campuses in Jerusalem. It provides care for approximately 70% of lymphoma patients diagnosed in Jerusalem. As part of a general strategy of quality assurance, a clinical audit of outcome of lymphoma patients diagnosed between 1987 and 1992 was carried out. In order to ensure coverage of all lymphoma cases cared for in multiple settings, multiple sources were used to ascertain lymphoma cases who had been cared for in our institution (O. Paltiel, submitted for publication). These sources included: ICD-9 codes for lymphoma from a computerized database of hospitalizations, outpatient database for the oncology clinics, and the Israel Cancer Registry for patients reported from Hadassah Hospital. Patients ascertained from the latter source were excluded if their registration number was unknown in the Hadassah computerized database. Demographic characteristics including age, sex, country of birth, religion, and coded place of residence were retrieved from the hospital’s

O. Paltiel et al.

94

administrative database, and confirmed using the unique identity number with the Population Registry of the Ministry of the Interior. The place of residence was defined as the Jerusalem district if it was within the confines of the municipality or in the surrounding suburbs or settlements to which Jerusalem hospitals were the closest institutions. Patients defined as residing outside of the Jerusalem district (‘‘referred’’) included those whose place of residence was in a catchment area administratively ascribed by the Sick Funds to another district. Until 1992 Hadassah was the exclusive Bone Marrow Transplant Center in Israel. Any contact with the treatment facilities was defined as treatment at Hadassah, however the vast majority of patients had multiple contacts with the institution. During the study period, standard chemo- and radiotherapeutic protocols were in use in our institution.

All analyses were performed using SPSS for Windows (SPSS Inc., Chicago, IL). Survival data were confirmed for all patients non-differentially using the Death Registry of the Ministry of Interior. Survival was measured from date of diagnosis, whether or not the diagnosis was made in Hadassah. Categorical variables were compared using chisquare. Kaplan-Meier survival curves were generated and compared using the log-rank statistic. In the analysis of survival we excluded patients with mycosis fungoides as well as those with uncertain histology. A Cox proportional hazards model controlling for the effect of age, exclusive admission to the bone marrow transplant department (and disease subtype, in the case of NHL) was constructed to assess the independent contribution of patient source (i.e., Jerusalem or other catchment area) on survival.

ate grade non-Hodgkin’s lymphoma (NHL). There is overrepresentation of high grade lymphomas, and of the combination of intermediate and high grade lymphomas versus other histologies in the referred ( p 5 0.02 and p 5 0.015, respectively) compared with the local population. This may be partially attributed to the referral of Muslim populations (including West Bank Arabs) from outside Jerusalem, in whom high grade histologies are more prevalent [5]. Survival of the study population in shown in Figure 1. Survival at 5 years was 83% for local patients with Hodgkin’s disease and 69% for those from outside of Jerusalem (p 5 0.17 for difference in the two curves). Local patients with a confirmed histologic diagnosis of NHL had lower overall survival compared to referred patients (median 47 vs. 63 months), (p 5 0.05). Restricting the analysis to the younger age groups (,60 years old) (Figure 2) did not alter the difference in survival curves between local and referred patients with NHL (once again lower survival for local patients was observed, p 5 0.05), although for Hodgkin’s disease local patients had a statistically significant improved survival compared to the referred group (p 5 0.02). Excluding patients whose sole hospitalization was in the BMT-D (Figure 3) diminished the differences in survival between Jerusalem and non-Jerusalem NHL patients, but a statistically non-significant difference in survival for Hodgkin’s disease patients, in favor of local residents remained (p 5 0.10). In the Cox model for NHL which included age, exclusive admission to BMT-D and disease subtype area of residence was not significantly associated with survival. For Hodgkin’s disease, controlling for age, the independent effect of area was confirmed, however when admission to the BMT-D was added to the model its effect was diminished (hazard ratio 0.54, p 5 0.07).

RESULTS

DISCUSSION

During the period of the audit, 716 patients with a diagnosis of lymphoma were ascertained from the three sources noted above (82% from the hospital ATD computer database). Of these, 399 (56.2%) were Jerusalem district residents (‘‘local’’) and 311 (43.8%) lived outside of the city (‘‘referred’’). Place of residence was missing for five patients. A comparison of demographic characteristics between these two groups is noted in Table 1. Referred patients were younger ( p , 0.0001), were more likely to have been born in Israel ( p , 0.0001), and to be non-Jewish ( p 5 0.004). Gender was not associated with referral. Referred cases were more likely to have been admitted to the Bone Marrow Transplantation Department (BMT-D) and more likely to have had an exclusive admission to the BMT-D. The distribution of histologic diagnoses is shown in Table 2 [4]. Overall, the distribution differs among the two groups ( p 5 0.01 for comparison of all diagnostic categories between the two groups), and specifically when comparing the prevalence of aggressive disease, such as high and intermedi-

Our audit revealed that survival for lymphoma patients treated in our institution was comparable to published results from the world literature [6–8]. We have shown significant differences in demographic and clinical characteristics between local and referred patients in our clinical audit of lymphoma in a teaching hospital. Patients who came from a distance, were referred, or chose not to attend their local hospital were younger and were less likely to be immigrants. In addition, they were more likely to have diseases requiring intensive therapy (high and intermediate grade NHL). Although referred or distant patients with NHL tended to be those with more aggressive disease, their survival was better than that of local patients. Given that one reason for referral to our center was BMT, a modality not available at other centers during the study period, we examined our local and referred population including and excluding those for whom the only contact with the institution was BMT. Excluding these patients diminished the survival gain of referred NHL patients. For Hodgkin’s disease, local

STATISTICAL ANALYSIS

Referral Bias in Audit of Lymphoma

95

TABLE 1. Comparison of demographic characteristics between Jerusalem residents and

others Residence a Characteristic Age 0–14 15–59 601 Gender Male Female Religion Jewish Other Country of birth Israel Other Admission to bone marrow transplant department Yes No Exclusive admission to bone marrow transplant department Yes No Total a

Other n (%)

p-Value

24 (6.0) 182 (45.8) 191 (48.1)

33 (10.2) 223 (69.3) 66 (20.5)

,0.0001

205 (51.6) 192 (48.4)

183 (56.8) 139 (43.2)

0.17

353 (88.9) 44 (11.1)

262 (81.4) 60 (18.6)

0.004

164 (41.3) 225 (56.7)

190 (59.0) 128 (39.8)

,0.0001

21 (6.6) 298 (93.4)

90 (34.2) 174 (65.8)

,0.0001

8 (2.5) 311 (97.5) 397 (55.2)

65 (24.6) 199 (75.4) 322 (44.8)

,0.0001

Five patients (0.7%) were missing data regarding place of residence.

patients continued to show a superior 5-year survival compared to referred patients. Adding exclusive admission to the BMT-D to the Cox model diminished the effect of catchment area in both disease categories. Thus, some of the effects of referral noted in this study may be confounded by the referral of patients for a specific procedure—BMT. Our study has several limitations. Important baseline prognostic factors such as stage, biochemical parameters such as lactate dehydrogenase levels (LDH) as well as performance status were not available using our sources, and we have no way of knowing whether or not they were differentially distributed among referred and local patients. However, in order for this to affect the observed survival results TABLE 2. Histologic classification a by place of residence

Residence Diagnosis Hodgkin’s disease Non-Hodgkin’s lymphoma Low grade/CLL b Intermediate grade High grade Mycosis fungoides Undefined NHL Total a

Jerusalem n (%)

Jerusalem n (%)

Other n (%)

113 (28.5)

83 (25.8)

78 104 17 15 70 397

(19.7) (26.2) (4.3) (3.8) (17.6) (55.2)

By the working formulation [4]. b CLL 5 chronic lymphocytic leukemia.

59 96 29 2 53 322

(18.3) (29.8) (9.0) (0.6) (16.5) (44.8)

one would have to presume worse prognostic features in referred patients with Hodgkin’s disease and more favorable ones in those with NHL, an unlikely interaction. In addition, only 70% of Jerusalem lymphoma cases are cared for at Hadassah, and the extent to which differential referral patterns take place among local residents is not known. Furthermore, the extent to which care was received at our institution exclusively is not discernable using our database. Nevertheless, important differences between the local and distant sub-populations were found and warrant discussion. Systematic differences in demographic characteristics between referred and local patients have been reported in the literature. In most series which have examined the issue, referral patients were younger [3,9–11]. In our series as well, referral patients were younger, despite the fact that the Jerusalem population is younger than that of the rest of the country [12]. There is no consistency with regard to gender differences; some studies have found that referred patients were more likely to be male [3,11], while others have found over-representation of women [9,10]. We did not find a sexpredilection. Other socioeconomic factors such as marital status, income, and education have been examined by other authors with inconsistent findings [10,13]. Severity of illness parameters have been compared in referral and local populations, and comparing community and teaching hospitals. Berman and colleagues [14] reported that the distribution of disease severity according to two

96

O. Paltiel et al.

FIGURE 1. Survival of lym-

phoma patients diagnosed 1987–1992 by place of residence and diagnosis. The top two curves represent Hodgkin’s disease patients, the bottom two represent patients with non-Hodgkin’s lymphoma. Censored patients are represented by tick marks or diamonds; catchment area: Jerusalem (r), other (1).

FIGURE 2. Survival of lym-

phoma patients diagnosed 1987–1992 by place of residence and diagnosis, excluding patients aged 60 years or older. Censored patients are represented by tick marks or diamonds; catchment area: Jerusalem (r), other (1).

Referral Bias in Audit of Lymphoma

97

FIGURE 3. Survival of lym-

phoma patients diagnosed 1987–1992 by place of residence and diagnosis, excluding patients whose only admission was to the Bone Marrow Transplant Department. Censored patients are represented by tick marks or diamonds; catchment area: Jerusalem (r), other (1).

scales demonstrated a greater proportion of severely ill patients admitted to teaching hospitals as compared with community hospitals. Hospital costs are higher for distant versus local patients [15], suggesting that distance may serve as a proxy for severity. While the number of co-morbid conditions [11] or tumor stage [16] may be higher in referred patients, functional status (possibly a more important indicator of severity) may actually be similar or even superior in the referred group [9]. This latter point may be the key element in explaining the paradox in the literature regarding referral bias. The paradox is that although referred patients are more complex, or have more advanced or aggressive disease, their prognosis may often be better than that of local patients. This has been found with respect to postoperative mortality [3,11,17]. The paradox suggests that referral bias is more complex than initially perceived and actually works in both directions: patients are defined as severe according to disease-specific criteria but may have ‘‘survival advantages’’ such as better functional status, or increased personal and socioeconomic resources to be able to travel or be referred [18]. This bi-directional bias may influence the literature in a number of ways. The predominance of complex or severe cases in a referral center may bias reports of the natural history of disease or co-prevalence of other conditions reported from tertiary centers [3,19] as evidenced by Sackett’s original example of secondary causes of hypertension at the Cleveland Clinic [20]. In addition, there is evidence that

more intensive investigations performed in tertiary centers result in classification or staging of patients as more advanced [16]. Conversely, reports on the efficacy or success rate of treatment modalities from referral centers may be biased toward improved survival or lower mortality if they include a large proportion of distant patients who were well or mobile enough to undergo the procedure. Furthermore, if patients travel from a distance there may be a temptation to intervene immediately (‘‘a bird in the hand’’), whereas treatment for local patients may be delayed. In our study we were unable to distinguish self-referral from physician referral, a factor which may also be associated with severity and differences in outcome [21]. When faced with a choice of care facilities for a particular medical problem, considerations such as distance and travel time influence patient preference [22] as do specialization of medical facilities [23]. In the case of lymphoma a sophisticated procedure (BMT) was only available in our institution, but patients admitted for this procedure accounted for only 34% of those referred. Patients may be drawn to a hospital with a given level of sophistication even when they themselves are not candidates for a sophisticated procedure. They may perceive a higher quality of care in centers where these procedures or technologies are available. CONCLUSION We have shown that patients referred to our center for the care of lymphoma are younger but have more aggressive his-

O. Paltiel et al.

98

tologies, while their survival may be worse, as good as, or better than local patients. Some, but not all, of the differences between referred and local patients can be accounted for by access to BMT in our institution exclusively. The paradox of increased complexity of referred patients yet equivalent or improved survival corresponds to findings in the literature. Since place of residence, or patient source (i.e., within or outside the usual hospital catchment area) may indeed represent a prognostic feature, it should be included as an important variable in accounts of natural history. When comparisons of treatment results are made between institutions, the catchment populations should also be described. The authors gratefully acknowledge the assistance of Lois Gordon in the preparation of the figures.

10. 11.

12. 13.

14. 15.

References 1. Sackett DL. Bias in analytic research. J Chron Dis 1979; 32: 51–63. 2. Redfield MM, Gersh BJ, Bailey KR, Ballard DJ, Rodehoffer RJ. Natural history of idiopathic dilated cardiomyopathy: Effect of referral bias and secular trend. J Am Coll Cardiol 1993; 22: 1921–1926. 3. Melton LJ. Selection bias in the referral of patients and the natural history of surgical conditions. Mayo Clin Proc 1985; 60: 880–889. 4. Rosenberg SA. National Cancer Institute sponsored study of classifications of non-Hodgkin’s lymphomas. Cancer 1982; 49: 2112–2135. 5. Paltiel O. Some aspects of lymphoma in Jerusalem and Israel. [Abstract] Middle East Cancer Society. Tel Aviv: International Regional Scientific Symposium. November 27–30, 1995; 15–16. 6. Fisher RI, Gaynor ER, Dahlberg S, et al. Comparison of a standard regimen (CHOP) with three intensive chemotherapy regimens for advanced non-Hodgkin’s lymphoma. N Engl J Med 1993; 328: 1002–1006. 7. Canellos GP, Anderson JR, Propert KJ, et al. Chemotherapy of advanced Hodgkin’s disease with MOPP, ABVD or MOPP alternating with ABVD. N Engl J Med 1992; 327: 1478– 1484. 8. Rohatiner AZS, Lister TA. Myeloablative therapy for follicular lymphoma. Hem/Onc Clin N Am 1991; 5: 1003–1012. 9. Greenberg ER, Dain BJ, Freeman DH, Yates JW, Korson R, Redfield MM. Referral of lung cancer patients to university

16.

17.

18. 19. 20. 21.

22. 23.

hospital cancer centers. A population-based study in two rural states. Cancer 1988; 62: 1647–1652. Nelson LM, Franklin GM, Hamman RF, Boteler DL, Baum HM, Burks JS. Referral bias in multiple sclerosis research. J Clin Epidemiol 1988; 44: 187–192. Warner MA, Hosking MP, Lobdell CM, Offord KP. Melton LJ III. Effects of referral bias on surgical outcomes: A populationbased study of surgical patients 90 years of age or older. Mayo Clin Proc 1990; 65: 1185–1191. Central Bureau of Statistics. Statistical Abstract of Israel 1994. No. 45. Jerusalem, Israel: Central Bureau of Statistics, 61. Greenberg ER, Chute CG, Stukel T, Baron JA, Freeman DH, Yates JW, Korson R. Social and economic factors in the choice of lung cancer treatment. A population-based study in two rural states. N Engl J Med 1988; 318: 612–617. Berman RA, Green J, Kwo D, Safian K, Botnick L. Severity of illness and the teaching hospital. J Med Education 1986; 61:1–9. Welch WG, Larson EB, Welch WP. Could distance be a proxy for severity-of-illness? A comparison of hospital costs in distant and local patients. HSR 1993; 28:441–458. Greenberg ER, Baron JA, Dain BJ, Freeman DH, Yates JW, Korson R. Cancer staging may have different meanings in academic and community hospitals. J Clin Epidemiol 1991; 44: 505–512. Ballard DJ, Bryant SC, O’Brien PC, Smith DW, Pine MB, Cortese DA. Referral selection bias in the Medicare Hospital Mortality Prediction Model: Are centers of referral for Medicare beneficiaries necessarily centers of excellence? HSR 1994; 28:771–784. Schneider AJ, Derr J. Referral bias in elderly patients. Mayo Clin Proc 1990; 65: 1272–1273. (Editorial) Maron BJ, Spirito P. Impact of selection biases on the perception of hypertrophic cardiomyopathy and its natural history. Am J Cardiol 1993; 72: 970–972. (Editorial) Gifford RW. Evaluation of the hypertensive patient with emphasis on detecting curable causes. Millbank Mem Fund Quart 1969; 47: 170–186. Paul RI, Christoffel KK, Binns HJ, Jaffe DM. Foreign body ingestions in children: Risk of complication varies with site of initial health care contact. Pediatric Practice Research Group. Pediatrics 1993; 91(1): 121–127. McGuirk MA, Porell FW. Spatial patterns of hospital utilization: The impact of distance and time. Inquiry 1984; 21: 84– 95. Mayer JD. The distance behavior of hospital patients: A disaggregated analysis. Soc Sci Med 1983; 17: 819–827.