- Email: [email protected]
Influence of clinician workload and patterns of treatment on survival from breast cancer
Influence of clinician workload and patterns of treatment survival from breast cancer
Summary
Introduction
Chemotherapy and hormone therapy prolong disease-free and overall survival for patients with breast cancer in the clinical-trial setting, but it is not clear if this translates into a benefit on a population basis. It is also not clear if surgical caseload has any influence on survival. We used cancer-registry data from 12861 patients with breast cancer treated in Yorkshire, UK, between 1979 and 1988, and found that patients of surgeons with higher rates of usage of chemotherapy and hormone therapy (regional mean usage 9·3%, range 0-46%) had prolonged survival. There was considerable variation in survival of breast cancer patients between surgeons, but their rate of use of chemotherapy and hormone therapy explained about 26% of this survival variation. Had the practice of the surgeons with the better outcomes been used by all treating clinicians, 5-year survival would have increased by about 4-5%. Examination of differences in survival as a function of consultant caseload demonstrated poorer results amongst those surgeons treating less than 30 new cases of breast cancer per year (risk ratio [95% Cl] for treating >30 compared with <10=0·85 [0·77-0·93]). We recommend that patients with breast cancer be dealt with only by clinicians who see more than 30 new cases per year and who have a full range of treatment options available within a multidisciplinary setting.
Reports
Lancet 1995; 345: 1265-70 See Editorial page 1251
on
of variation in survival for patients with breast between regions and countries have attracted attention,’ but attempts to explain such differences are few and unsatisfactory. It is not clear which factors contribute to these differences, although stage at presentation and treatments administered might be expected to be important. Cultural, dietary, and environmental factors (often unspecified) are cited but with little supporting evidence. There has also been interest in whether the caseload of an individual clinician has any bearing on outcome for patients with breast cancer. Treatment variations between surgeons have been identified,2 and these differences in clinical practice might account for some variation in survival. The relation between increased caseload and improved outcome has been shown for patients with rarer cancers such as childhood leukaemia3,4 or adults with oesophageal cancer,5 but has not been studied for patients with breast cancer. Cancer registries are a potentially valuable source of data for examination of variations in survival because of the many cases recorded and the long-term nature of these records, which facilitates outcome analysis. Treatment information is now recorded routinely by all cancer registries (from July, 1993) as part of the minimum data set required by the UK Office of Population Censuses and Surveys-the body with overall responsibility for national cancer statistics. The Yorkshire cancer registry has recorded data on treatment given within 9 weeks of diagnosis for many years. The use of this registry data for retrospective analysis and audit is, therefore, possible, although quality assurance of such data is important. The ability of the data to discriminate case-mix is somewhat compromised by the absence of clinical-staging information. This problem can be cancer
compensated for, to some extent, by analytical techniques such as linear-regression analysis. We have examined variation in 5-year survival between
Department of Surgery, Royal Infirmary, Huddersfield HD3 3EA, UK (R Sainsbury FRCS); and Department of Cancer Studies (Prof R Haward FFPHM), Clinical Trials and Research Unit (C Round MSC), and Yorkshire Cancer Registry (L Rider BA, C Johnston MSc), Yorkshire Cancer Organisation, Cookridge Hospital, Leeds LS16 6QB, UK Correspondence to: Mr Richard Sainsbury or Professor Bob Haward
health districts and between individual surgeons within the Yorkshire Regional Health Authority (population 3-6 million). We have attempted to explain differences by examining case-mix (including age of patient, disease extent at presentation, degree of deprivation, and caseload of treating consultant) and treatment regimens given to the patients. Linking treatment data with survival for individual clinicians can be used for audit and for analysis of the nature of clinical activity and survival within a population at large rather than for a selected group within a clinical trial. Linear-regression (with outcome as the dependent variable) was used to examine the degree to which variations in clinical practice might contribute to outcome.
The
basic adjusting the
proposition data
to
examined was-after account of age, social
we
take
1265
Table 1: Number by caseload
(%) of consultants and patients
deprivation, and the
in each caseload
of disease-are there be explained either by the individual surgeons or by
extent
differences in survival that numbers of cases treated their treatment patterns?
can
by
Patients and methods
Population The
population studied was all female residents in the Yorkshire Regional Health Authority area diagnosed with invasive breast cancer between 1979 and 1988 and treated by surgery of curative intent. This gave a study population of 12 861 patients. Statistical methods
Survival after treatment for breast cancer was examined in relation to the case-mix of the patients, factors linked to the patient directly, and factors related to the characteristics of the consultants managing the patients. The case-mix factors considered were age, extent of disease, tumour grade, and socioeconomic deprivation. Associations between the year of diagnosis, treatments given to patients, and survival were also examined. Factors characterising the consultants were clinical caseload and the percentage of each consultant’s patients who were treated by different forms of therapy. These factors were given the same value for all the patients assigned to a particular consultant, irrespective of the actual treatments given to the individual patients. Data were analysed by Cox’s proportional hazards regression.6 The associations of the explanatory variables with the survival
CT=chemotherapy, HT=hormone therapy, RT=radiotherapy. Table 2: Number (%) of patients in each tumour
1266
category, and of patients in each age group and disease category
were examined both individually and together. Using an approach outlined by Henderson,’ we calculated a measure of the proportion of variation in survival among patients explained by the factors considered. This is calculated as r=1-exp(-LR/n) where LR is the (partial) likelihood ratio statistic for testing Ho: b=0 against He no restrictions on b. The variation in patient
time
survival associated with differences among consultants was estimated by fitting a model containing-as well as any patient factors-a separate indicator for each consultant. An approximation to the proportion of variation among consultants attributable to each of the consultant factors was then estimated. When examining the risk ratios for the other consultant factors it should be borne in mind that these were estimates for a 10% difference in practice. A 20% difference would double the normal logarithm of the risk ratio.
Survival time was measured from the date of the first treatment the date of death. Analysis was based on deaths from all causes, not just those due to breast cancer. For those patients still alive at the time of writing, the censoring date was taken to be January 1, 1995. The cancer registry does not actively follow up cases so there might be a small number of patients who have emigrated and for whom death details are not available. to
grouped into four age bands: those under 50, 80 and over. Each patient was put into one of and 50-64, 65-79, three disease categories: no nodal involvement or metastases recorded, nodal involvement but no metastases, and those with metastases. The first group of patients might include some with a poor prognosis, possibly with nodal involvement or metastases, whose cancers were not investigated fully. Tumours were graded Patients
were
grade, deprivation category, and treatment category by case load
poorly
each period
well differentiated, moderately differentiated, poorly differentiated, or unknown. Tumour grade was known in 53% of as
Townsend
index of affluence/deprivation8 was data for every ward in the region. Every patient was assigned the deprivation score of her ward of residence. The wards were divided into five categories according to their deprivation score where category I was the most affluent 20% of wards and category V the most deprived 20%. In 1991, 10-0% of women lived in the most affluent wards and 31-1% in the most deprived. To test whether the survival of breast cancer patients improved over the study period, the year of diagnosis was recorded and the interval 1979-1988 divided into five 2-year periods. Patients were divided into four groups according to the treatments given in the first 9 weeks of therapy: surgery alone, surgery plus chemotherapy and/or hormone therapy, surgery plus radiotherapy, and surgery plus radiotherapy and either chemotherapy and/or hormone therapy. The association of factors characterising the consultants with differences in the survival of their patients was examined. The caseload of each consultant was estimated to be the median number of patients seen as primary consultations per year, calculated over the period from the first registration for that consultant to the last registration. Patients who did not receive surgery were included in this calculation because our interest centred on the management of patients rather than the number of operations done by each consultant. Data for the years 1976-1992 were used in the calculation to allow for consultants who ended their practice early in the period under study, or began late in the period. The caseloads thus calculated were categorised into less than 10 patients per year, 20-29 patients, 30-49 patients, and 50 or more patients per year. For each of the consultants in the study, the percentage of their patients treated by mastectomy (as opposed to local excision), radiotherapy, chemotherapy, hormone therapy, or surgery alone was calculated. All consultants with a caseload of less than 10 patients per year were considered together in these calculations. Every patient was assigned to the caseload category and the therapy percentages of her primary consultant. cases.
The
calculated from 1991
census
Results Case-mix The 12 861 patients in the study were managed by 180 consultants. 60 consultants treated 10 or more patients per year; these were all surgeons. The 8 surgeons who treated 50 or more patients per year dealt with 30% of the patients. Those consultants managing less than 10 patients per year treated 10% of the cases (table 1). The mean age of patients was slightly higher for those consultants with a caseload of less than 10 patients per year (60-11 years, overall mean 59-0 years). Those surgeons seeing 30 or more patients saw slightly younger patients (table 1). The percentage of patients in each disease category did not vary greatly among caseload categories (table 1). About 4-5% of tumours were well differentiated in each of the caseload categories (8% of those with known grade). A lower proportion of grades were categorised as unknown for the caseload category 30-49, with a corresponding increase in patients with
or
moderately differentiated
tumours
(table 2).
consistent differences in the proportions of patients living in deprived areas among caseload categories. Overall, 30% of patients lived in the most deprived electoral wards (table 2). Those consultants managing 30 or more patients per year gave combined therapy (surgery with radiotherapy and chemotherapy or hormone therapy) to a higher percentage of their patients (24%) than other consultants. 42% of patients managed by consultants in the lowest caseload category and 38% in the 10-29 category were treated by surgery alone (table 2). Over the period under study, there was a trend to increasing numbers of patients being treated by consultants with higher caseloads. During 1979-1980, 41 % of patients were treated by consultants managing 30 or more patients per year; this figure had risen to 51 % by 1987-1988 (table 3). The proportion of patients receiving each treatment method in the 9 weeks after diagnoses varied considerably among individual surgeons. Of those with a caseload of 10 or more patients per year, the mastectomy rate varied between 22 and 98%, the radiotherapy rate between 4 and 82%, the chemotherapy rate between 0 and 46%, and the hormone therapy rate between 0 and 86% (table 4).
There
were no
Survival
analysis Overall, 54% of the patients diagnosed in the period under study had died by the beginning of 1995. Table 5 gives the results of fitting survival models with the various explanatory factors. Each factor was considered in isolation, then all the case-mix factors were included together with the other patient factors, and consultant factors were examined individually after allowing for casemix. Finally, all the case-mix and patient factors were included together and each consultant factor was examined individually after allowing for case-mix and the other patient factors. The largest differences in survival were due to the casemix. Age, extent of disease, and, to a lesser degree, grade, all showed the expected trend in prognosis. The survival of patients with unknown grade of tumour was intermediate between those with moderately and poorly differentiated tumours. There was little difference in survival between patients resident in the more affluent wards and those resident in wards of average or below average affluence. However, patients in the most deprived wards had a higher risk ratio (1-16, 95% CI 1-1-1-22) than the rest of the population. There were significant differences in survival among patients in the four treatment categories. Those patients receiving radiotherapy and either chemotherapy or hormone therapy had the best survival. When treatment was examined in isolation, the survival of patients treated by surgery alone was comparable with that of those treated by radiotherapy plus chemotherapy or hormone
Table 4: Summary of consultant factors (% of receiving therapy) for individual surgeons
patients
1267
CT=chemotherapy, HT=hormone therapy, RT=radiotherapy, S=surgery. Table 5: Risk ratios (95% CI) of survival
therapy, and patients treated by either chemotherapy or hormone therapy had poorer survival. However, when case-mix factors were accounted for, survival of the three groups (surgery alone, surgery plus radiotherapy, and surgery plus chemotherapy or hormone therapy) was all similarly worse than that of the group receiving the most intensive management. There was a gradual overall improvement in survival over the study period. This improvement was largely explained by the other factors considered above, but survival in 1987-1988 was significantly better than that in the earlier years (risk ratio relative to 1979-1980=0-91,
chemotherapy and hormone therapy
were
significantly
associated with a reduced risk ratio and the percentage of patients treated by surgery alone was associated with a slightly raised risk. After adjustment for the case-mix and other patients factors, these factors remained statistically significant and use of mastectomy also seemed associated with a slightly reduced risk. Of all the overall factors, use of chemotherapy had the greatest association with a reduction in risk (relative risk 0-95, 0-93-0-97). After controlling for the case-mix and other patient factors, variation among consultants accounted for about 8% of the explainable differences in survival. Table 6
0-83-0-99). There was no difference in survival between patients treated by consultants with caseloads of less than 10 patients per year and those managed by surgeons with caseloads between 10 and 29 patients per year. Patients treated by surgeons with caseloads of more than 29 patients per year had significantly better survival. The difference was reduced slightly when case-mix and treatment were allowed for in the analysis but was still significant (risk ratio 0-85, 0-77-0-93, compared with the less than 10 category). Considering other consultant factors individually, without adjustment for case-mix or the treatment given, 1268
Table 6: Percentage variation in survival attributable to differences among consultants that is accounted for by consultant factors
Table 7: Expected 5-year survival percentages associated with different levels of use of chemotherapy by consultants
shows the percentage of this variation among consultants explained by each of the consultant factors examined. The factors most clearly associated with differences in survival results among consultants were caseload and use of chemotherapy, which both accounted for between 20 and 25% of the variation in results Expected survival percentages at 5 years associated with varying use of chemotherapy by consultants are shown in table 7. All other factors were kept at their mean levels and the figures shown were obtained after adjustment for case-mix, period, and actual treatment given. An 10% increase in use of chemotherapy was associated with about a 1-4% increase in 5-year survival.
Discussion The initial aims of this study were to examine whether or not differences in survival from breast cancer could be explained by variations in treatment regimens and surgeons’ caseload. The fulfilment of these aims raised questions as to the suitability of cancer-registry data for such an analysis. Part of the analysis described has, therefore, concentrated on making allowance for the most influential of the case-mix variables-namely, age, stage of disease, and social deprivation. Although these are important variable at an individual level, and in table 5 show predictable effects on outcome (substantially greater risk in older women and those with more advanced disease, and a slightly greater risk in the most socially disadvantaged), their influence on the results of multivariate analysis turns out to be small. Thus there was little change in the risk ratios relating caseload to survival whether or not a range of such factors were included in the analysis. How far can variation in survival be explained by patterns of treatment and how far does it remain unaccounted for? Can any of the differences be explained by consultant caseload? There is a great deal of current interest in moves to concentrate cancer treatments in the hands of fewer individuals with more specialist training. This trend has been shown to be associated with improved survival for a range of cancers.9 Reviewers have concluded that there is a relation between number of patients treated and outcome,10-12 but they are less clear about quantitating the relation except in very limited cases (such as coronary-artery-bypass grafting in the
USA). In this
study,
one
factor related
to
survival
was
clinical
caseload-ie, annual number of new patients with invasive breast
by the consultant. The difference in survival between patients treated by consultants patient with lower and higher caseloads was not explained by other factors, including type of treatment. When caseload was considered alone, patients managed by consultants treating more than 30 new patients per year fared better than those treated by consultants with lower caseloads. cancer seen
Further subdivision of the caseload categories would have resulted in loss of statistical power. We were unable to address the impact of higher caseloads because there were only a few surgeons with large caseloads in our survey. However, a clear relation was established between caseload and survival with a threshold for better outcome of 30 new patients per year. The second part of the study compared average survival among breast cancer patients treated by individual consultant surgeons. The aim was to gain an understanding of the impact of variations in management strategies on patient survival. The most important explanatory variable found was the level of chemotherapy used by the consultant, with higher levels linked with increased survival. After allowing for case-mix and other patient variables, 20% of the variation in survival was explained by chemotherapy alone and 6% by hormone therapy alone. These figures also suggest that specific treatment variations are important explanatory factor for the differences between survival curves for individual surgeons. Indeed, almost a quarter (26%) of the statistical variation in this measurement was accounted for by differences in the use of chemotherapy and hormone
therapy. Treatment data (up to 9 weeks after diagnosis) have been recorded by the Yorkshire Cancer Registry for many years but did not take account of qualitative variations in the agents given, their dosages, and the protocols that may have been in use at the time of data collection. This fact will lead to a bias in underestimating the relation between treatments and outcome. There are likely to have been many suboptimal treatments given over the years in question. In addition, the problems of using a longterm database in which changes in practice occur with time (as well as retirements and new appointments) must not be minimised. The benefits of chemotherapy and hormone therapy may, therefore, have been understated although we can offer no evidence for the magnitude of this effect. The clear relation between these two therapies and outcome is consistent with the many studies that have shown outcome benefits from these interventions. Such evidence arises from clinical trials under controlled conditions and it is important to see that this can be confirmed on a population basis. Another explanation for the significance of the links between use of chemotherapy and hormone therapy, case load, and survival might lie in the value, now widely accepted, of a multidisciplinary approach to the management of breast cancer. This has the objective of ensuring that all relevant treatment methods that have a place in modern management are properly considered and deployed when necessary. Higher rates of these treatments, taken together, are likely to characterise services working efficiently. Thus some of the improved survival associated with greater consultant caseload might be due to those consultants who treat more patients also making more use of the full range of therapeutic options to the benefit of their patients. It is tempting to estimate the impact on survival that might result from the widespread adoption of higher levels of chemotherapy and hormone therapy. Had chemotherapy been used for 40% of patients (instead of the mean regional usage of 9-3%), the calculations in table 7 predict an improvement in 5-year survival of about 4%, and for hormone therapy (already much more widely adopted) the improvement in survival lies between 1 and 1269
1-5%. These percentages approximate to an additional 500 women still alive-a worthwhile gain. There was no relation between rates of use of radiotherapy and survival. This finding might have been anticipated because the primary role of radiotherapy in breast cancer lies in local control. High rates of surgery alone were significantly associated with a poorer survival, an observation lending support to the belief that multiple therapies, correctly deployed, improve survival. Surgeons with higher caseloads but lower use of multiple therapies are exceptions to the general finding of an improved survival for increased caseload. This is an important confirmatory observation to the suggested interpretation of the results for chemotherapy and hormone therapy. The effect of number of patients treated on outcome of breast cancer may not, therefore, as might be argued for some gastrointestinal cancers, be a simple function of the skill of the surgical team, but is rather a function of clinical organisation. The volume effect reflects the ability to bring together the necessary clinical disciplines and expertise across the full therapeutic range. Gillis et al13 considered the issue of multidisciplinary practice to be relevant to the interpretation of their data on survival from ovarian cancer. It has also been used to explain at least part of the difference in survival for Hodgkin’s disease between centres and peripheral hospitals in the USA. 14 The report on cancer servicesl5 and guideline documents on breast cancer" emphasise moves to improve breast cancer services based on the two principles of surgical site specialisation with higher workloads and multidisciplinary working. Pressure groups are urging women to look for services with these characteristics." Many such changes in services are under way with widespread acceptance of their desirability. These changes are consistent with the evidence on caseload and treatment variations described in this paper.
1270
References 1
2
3 4
5
6 7 8 9
Berino F, Sant M, Verdecchia A, Capocaccia R, Hakulinnen T, Esteve J, eds. Survival for cancer patients in Europe. IARC Sci Publ 132
(in press). Sainsbury JRC,
Rider L, Smith A, McAdam WFA. Does it matter where you live? Treatment variation for breast cancer in Yorkshire. Br J Cancer 1995; 71: 1275-78. Stiller CA. Centralisation of treatment and survival rates for cancer. Arch Dis Child 1988; 63: 23-30. Kramer S, Meadows AT, Pastor G, et al. Influence of place of treatment on diagnosis, treatment and survival in three paediatric solid tumours. J Clin Oncol 1984; 2: 917-23. Matthews HR, Powell DJ, McConkey CC. Effects of the results of surgical experience on the results of resection for oesophageal carcinoma. Br J Surgery 1986; 73: 621-23. Cox DR. Regression models and life tables (with discussion). J R Stat Soc 1972; 34: 187-220. Henderson R. Problems and prediction in survival-data analysis. Stat Med 1955; 14: 161-84. Morphet C. Interpretation of small area census data-area 24/1. London: HM Stationery Office, 1992: 62-72. McArdle CS, Hole D. Impact of variability among surgeons on postoperative morbidity and mortality and ultimate survival. BMJ
1991; 302: 150l-05. 10 Stiller CA. Centralised treatment, entry to trials and survival. Br J Cancer 1994; 70: 352-62. 11 Luft HS, Garnick DW, Mark DH, McPhee SJ. Hospital volume, physician volume and patient outcomes. Ann Arbor: Health Administration Press, 1990. 12 Luft HS, Hunt SS, Maerk SC. The volume-outcome relationship: practice makes perfect or selective referral patterns? Health Services Res 1987; 22: 157-82. 13 Gillis CR, Hold DJ, Still RM, et al. Medical audit, cancer registration and survival in ovarian cancer. Lancet 1991; 337: 611-12. 14 Davis S, Dahlberg S, Myers M, et al. Hodgkin’s disease in the United States: a comparison of patient characteristics and survival in the centralised cancer patient data system and the surveillance, epidemiology and end results program. J Natl Cancer Inst 1987; 78: 471-78. 15 Consultative document: a policy framework for commissioning cancer services. London: Department of Health, 1994. 16 The provision of breast services in the UK: the advantages of specialist breast units. Report of a working party of the British Breast Group. Edinburgh: Churchill Livingstone, 1995. 17 The MacMillan directory of breast cancer services in the UK. London: Cancer Relief MacMillan Fund and The King’s Fund. 1995.
Summary
Introduction
Chemotherapy and hormone therapy prolong disease-free and overall survival for patients with breast cancer in the clinical-trial setting, but it is not clear if this translates into a benefit on a population basis. It is also not clear if surgical caseload has any influence on survival. We used cancer-registry data from 12861 patients with breast cancer treated in Yorkshire, UK, between 1979 and 1988, and found that patients of surgeons with higher rates of usage of chemotherapy and hormone therapy (regional mean usage 9·3%, range 0-46%) had prolonged survival. There was considerable variation in survival of breast cancer patients between surgeons, but their rate of use of chemotherapy and hormone therapy explained about 26% of this survival variation. Had the practice of the surgeons with the better outcomes been used by all treating clinicians, 5-year survival would have increased by about 4-5%. Examination of differences in survival as a function of consultant caseload demonstrated poorer results amongst those surgeons treating less than 30 new cases of breast cancer per year (risk ratio [95% Cl] for treating >30 compared with <10=0·85 [0·77-0·93]). We recommend that patients with breast cancer be dealt with only by clinicians who see more than 30 new cases per year and who have a full range of treatment options available within a multidisciplinary setting.
Reports
Lancet 1995; 345: 1265-70 See Editorial page 1251
on
of variation in survival for patients with breast between regions and countries have attracted attention,’ but attempts to explain such differences are few and unsatisfactory. It is not clear which factors contribute to these differences, although stage at presentation and treatments administered might be expected to be important. Cultural, dietary, and environmental factors (often unspecified) are cited but with little supporting evidence. There has also been interest in whether the caseload of an individual clinician has any bearing on outcome for patients with breast cancer. Treatment variations between surgeons have been identified,2 and these differences in clinical practice might account for some variation in survival. The relation between increased caseload and improved outcome has been shown for patients with rarer cancers such as childhood leukaemia3,4 or adults with oesophageal cancer,5 but has not been studied for patients with breast cancer. Cancer registries are a potentially valuable source of data for examination of variations in survival because of the many cases recorded and the long-term nature of these records, which facilitates outcome analysis. Treatment information is now recorded routinely by all cancer registries (from July, 1993) as part of the minimum data set required by the UK Office of Population Censuses and Surveys-the body with overall responsibility for national cancer statistics. The Yorkshire cancer registry has recorded data on treatment given within 9 weeks of diagnosis for many years. The use of this registry data for retrospective analysis and audit is, therefore, possible, although quality assurance of such data is important. The ability of the data to discriminate case-mix is somewhat compromised by the absence of clinical-staging information. This problem can be cancer
compensated for, to some extent, by analytical techniques such as linear-regression analysis. We have examined variation in 5-year survival between
Department of Surgery, Royal Infirmary, Huddersfield HD3 3EA, UK (R Sainsbury FRCS); and Department of Cancer Studies (Prof R Haward FFPHM), Clinical Trials and Research Unit (C Round MSC), and Yorkshire Cancer Registry (L Rider BA, C Johnston MSc), Yorkshire Cancer Organisation, Cookridge Hospital, Leeds LS16 6QB, UK Correspondence to: Mr Richard Sainsbury or Professor Bob Haward
health districts and between individual surgeons within the Yorkshire Regional Health Authority (population 3-6 million). We have attempted to explain differences by examining case-mix (including age of patient, disease extent at presentation, degree of deprivation, and caseload of treating consultant) and treatment regimens given to the patients. Linking treatment data with survival for individual clinicians can be used for audit and for analysis of the nature of clinical activity and survival within a population at large rather than for a selected group within a clinical trial. Linear-regression (with outcome as the dependent variable) was used to examine the degree to which variations in clinical practice might contribute to outcome.
The
basic adjusting the
proposition data
to
examined was-after account of age, social
we
take
1265
Table 1: Number by caseload
(%) of consultants and patients
deprivation, and the
in each caseload
of disease-are there be explained either by the individual surgeons or by
extent
differences in survival that numbers of cases treated their treatment patterns?
can
by
Patients and methods
Population The
population studied was all female residents in the Yorkshire Regional Health Authority area diagnosed with invasive breast cancer between 1979 and 1988 and treated by surgery of curative intent. This gave a study population of 12 861 patients. Statistical methods
Survival after treatment for breast cancer was examined in relation to the case-mix of the patients, factors linked to the patient directly, and factors related to the characteristics of the consultants managing the patients. The case-mix factors considered were age, extent of disease, tumour grade, and socioeconomic deprivation. Associations between the year of diagnosis, treatments given to patients, and survival were also examined. Factors characterising the consultants were clinical caseload and the percentage of each consultant’s patients who were treated by different forms of therapy. These factors were given the same value for all the patients assigned to a particular consultant, irrespective of the actual treatments given to the individual patients. Data were analysed by Cox’s proportional hazards regression.6 The associations of the explanatory variables with the survival
CT=chemotherapy, HT=hormone therapy, RT=radiotherapy. Table 2: Number (%) of patients in each tumour
1266
category, and of patients in each age group and disease category
were examined both individually and together. Using an approach outlined by Henderson,’ we calculated a measure of the proportion of variation in survival among patients explained by the factors considered. This is calculated as r=1-exp(-LR/n) where LR is the (partial) likelihood ratio statistic for testing Ho: b=0 against He no restrictions on b. The variation in patient
time
survival associated with differences among consultants was estimated by fitting a model containing-as well as any patient factors-a separate indicator for each consultant. An approximation to the proportion of variation among consultants attributable to each of the consultant factors was then estimated. When examining the risk ratios for the other consultant factors it should be borne in mind that these were estimates for a 10% difference in practice. A 20% difference would double the normal logarithm of the risk ratio.
Survival time was measured from the date of the first treatment the date of death. Analysis was based on deaths from all causes, not just those due to breast cancer. For those patients still alive at the time of writing, the censoring date was taken to be January 1, 1995. The cancer registry does not actively follow up cases so there might be a small number of patients who have emigrated and for whom death details are not available. to
grouped into four age bands: those under 50, 80 and over. Each patient was put into one of and 50-64, 65-79, three disease categories: no nodal involvement or metastases recorded, nodal involvement but no metastases, and those with metastases. The first group of patients might include some with a poor prognosis, possibly with nodal involvement or metastases, whose cancers were not investigated fully. Tumours were graded Patients
were
grade, deprivation category, and treatment category by case load
poorly
each period
well differentiated, moderately differentiated, poorly differentiated, or unknown. Tumour grade was known in 53% of as
Townsend
index of affluence/deprivation8 was data for every ward in the region. Every patient was assigned the deprivation score of her ward of residence. The wards were divided into five categories according to their deprivation score where category I was the most affluent 20% of wards and category V the most deprived 20%. In 1991, 10-0% of women lived in the most affluent wards and 31-1% in the most deprived. To test whether the survival of breast cancer patients improved over the study period, the year of diagnosis was recorded and the interval 1979-1988 divided into five 2-year periods. Patients were divided into four groups according to the treatments given in the first 9 weeks of therapy: surgery alone, surgery plus chemotherapy and/or hormone therapy, surgery plus radiotherapy, and surgery plus radiotherapy and either chemotherapy and/or hormone therapy. The association of factors characterising the consultants with differences in the survival of their patients was examined. The caseload of each consultant was estimated to be the median number of patients seen as primary consultations per year, calculated over the period from the first registration for that consultant to the last registration. Patients who did not receive surgery were included in this calculation because our interest centred on the management of patients rather than the number of operations done by each consultant. Data for the years 1976-1992 were used in the calculation to allow for consultants who ended their practice early in the period under study, or began late in the period. The caseloads thus calculated were categorised into less than 10 patients per year, 20-29 patients, 30-49 patients, and 50 or more patients per year. For each of the consultants in the study, the percentage of their patients treated by mastectomy (as opposed to local excision), radiotherapy, chemotherapy, hormone therapy, or surgery alone was calculated. All consultants with a caseload of less than 10 patients per year were considered together in these calculations. Every patient was assigned to the caseload category and the therapy percentages of her primary consultant. cases.
The
calculated from 1991
census
Results Case-mix The 12 861 patients in the study were managed by 180 consultants. 60 consultants treated 10 or more patients per year; these were all surgeons. The 8 surgeons who treated 50 or more patients per year dealt with 30% of the patients. Those consultants managing less than 10 patients per year treated 10% of the cases (table 1). The mean age of patients was slightly higher for those consultants with a caseload of less than 10 patients per year (60-11 years, overall mean 59-0 years). Those surgeons seeing 30 or more patients saw slightly younger patients (table 1). The percentage of patients in each disease category did not vary greatly among caseload categories (table 1). About 4-5% of tumours were well differentiated in each of the caseload categories (8% of those with known grade). A lower proportion of grades were categorised as unknown for the caseload category 30-49, with a corresponding increase in patients with
or
moderately differentiated
tumours
(table 2).
consistent differences in the proportions of patients living in deprived areas among caseload categories. Overall, 30% of patients lived in the most deprived electoral wards (table 2). Those consultants managing 30 or more patients per year gave combined therapy (surgery with radiotherapy and chemotherapy or hormone therapy) to a higher percentage of their patients (24%) than other consultants. 42% of patients managed by consultants in the lowest caseload category and 38% in the 10-29 category were treated by surgery alone (table 2). Over the period under study, there was a trend to increasing numbers of patients being treated by consultants with higher caseloads. During 1979-1980, 41 % of patients were treated by consultants managing 30 or more patients per year; this figure had risen to 51 % by 1987-1988 (table 3). The proportion of patients receiving each treatment method in the 9 weeks after diagnoses varied considerably among individual surgeons. Of those with a caseload of 10 or more patients per year, the mastectomy rate varied between 22 and 98%, the radiotherapy rate between 4 and 82%, the chemotherapy rate between 0 and 46%, and the hormone therapy rate between 0 and 86% (table 4).
There
were no
Survival
analysis Overall, 54% of the patients diagnosed in the period under study had died by the beginning of 1995. Table 5 gives the results of fitting survival models with the various explanatory factors. Each factor was considered in isolation, then all the case-mix factors were included together with the other patient factors, and consultant factors were examined individually after allowing for casemix. Finally, all the case-mix and patient factors were included together and each consultant factor was examined individually after allowing for case-mix and the other patient factors. The largest differences in survival were due to the casemix. Age, extent of disease, and, to a lesser degree, grade, all showed the expected trend in prognosis. The survival of patients with unknown grade of tumour was intermediate between those with moderately and poorly differentiated tumours. There was little difference in survival between patients resident in the more affluent wards and those resident in wards of average or below average affluence. However, patients in the most deprived wards had a higher risk ratio (1-16, 95% CI 1-1-1-22) than the rest of the population. There were significant differences in survival among patients in the four treatment categories. Those patients receiving radiotherapy and either chemotherapy or hormone therapy had the best survival. When treatment was examined in isolation, the survival of patients treated by surgery alone was comparable with that of those treated by radiotherapy plus chemotherapy or hormone
Table 4: Summary of consultant factors (% of receiving therapy) for individual surgeons
patients
1267
CT=chemotherapy, HT=hormone therapy, RT=radiotherapy, S=surgery. Table 5: Risk ratios (95% CI) of survival
therapy, and patients treated by either chemotherapy or hormone therapy had poorer survival. However, when case-mix factors were accounted for, survival of the three groups (surgery alone, surgery plus radiotherapy, and surgery plus chemotherapy or hormone therapy) was all similarly worse than that of the group receiving the most intensive management. There was a gradual overall improvement in survival over the study period. This improvement was largely explained by the other factors considered above, but survival in 1987-1988 was significantly better than that in the earlier years (risk ratio relative to 1979-1980=0-91,
chemotherapy and hormone therapy
were
significantly
associated with a reduced risk ratio and the percentage of patients treated by surgery alone was associated with a slightly raised risk. After adjustment for the case-mix and other patients factors, these factors remained statistically significant and use of mastectomy also seemed associated with a slightly reduced risk. Of all the overall factors, use of chemotherapy had the greatest association with a reduction in risk (relative risk 0-95, 0-93-0-97). After controlling for the case-mix and other patient factors, variation among consultants accounted for about 8% of the explainable differences in survival. Table 6
0-83-0-99). There was no difference in survival between patients treated by consultants with caseloads of less than 10 patients per year and those managed by surgeons with caseloads between 10 and 29 patients per year. Patients treated by surgeons with caseloads of more than 29 patients per year had significantly better survival. The difference was reduced slightly when case-mix and treatment were allowed for in the analysis but was still significant (risk ratio 0-85, 0-77-0-93, compared with the less than 10 category). Considering other consultant factors individually, without adjustment for case-mix or the treatment given, 1268
Table 6: Percentage variation in survival attributable to differences among consultants that is accounted for by consultant factors
Table 7: Expected 5-year survival percentages associated with different levels of use of chemotherapy by consultants
shows the percentage of this variation among consultants explained by each of the consultant factors examined. The factors most clearly associated with differences in survival results among consultants were caseload and use of chemotherapy, which both accounted for between 20 and 25% of the variation in results Expected survival percentages at 5 years associated with varying use of chemotherapy by consultants are shown in table 7. All other factors were kept at their mean levels and the figures shown were obtained after adjustment for case-mix, period, and actual treatment given. An 10% increase in use of chemotherapy was associated with about a 1-4% increase in 5-year survival.
Discussion The initial aims of this study were to examine whether or not differences in survival from breast cancer could be explained by variations in treatment regimens and surgeons’ caseload. The fulfilment of these aims raised questions as to the suitability of cancer-registry data for such an analysis. Part of the analysis described has, therefore, concentrated on making allowance for the most influential of the case-mix variables-namely, age, stage of disease, and social deprivation. Although these are important variable at an individual level, and in table 5 show predictable effects on outcome (substantially greater risk in older women and those with more advanced disease, and a slightly greater risk in the most socially disadvantaged), their influence on the results of multivariate analysis turns out to be small. Thus there was little change in the risk ratios relating caseload to survival whether or not a range of such factors were included in the analysis. How far can variation in survival be explained by patterns of treatment and how far does it remain unaccounted for? Can any of the differences be explained by consultant caseload? There is a great deal of current interest in moves to concentrate cancer treatments in the hands of fewer individuals with more specialist training. This trend has been shown to be associated with improved survival for a range of cancers.9 Reviewers have concluded that there is a relation between number of patients treated and outcome,10-12 but they are less clear about quantitating the relation except in very limited cases (such as coronary-artery-bypass grafting in the
USA). In this
study,
one
factor related
to
survival
was
clinical
caseload-ie, annual number of new patients with invasive breast
by the consultant. The difference in survival between patients treated by consultants patient with lower and higher caseloads was not explained by other factors, including type of treatment. When caseload was considered alone, patients managed by consultants treating more than 30 new patients per year fared better than those treated by consultants with lower caseloads. cancer seen
Further subdivision of the caseload categories would have resulted in loss of statistical power. We were unable to address the impact of higher caseloads because there were only a few surgeons with large caseloads in our survey. However, a clear relation was established between caseload and survival with a threshold for better outcome of 30 new patients per year. The second part of the study compared average survival among breast cancer patients treated by individual consultant surgeons. The aim was to gain an understanding of the impact of variations in management strategies on patient survival. The most important explanatory variable found was the level of chemotherapy used by the consultant, with higher levels linked with increased survival. After allowing for case-mix and other patient variables, 20% of the variation in survival was explained by chemotherapy alone and 6% by hormone therapy alone. These figures also suggest that specific treatment variations are important explanatory factor for the differences between survival curves for individual surgeons. Indeed, almost a quarter (26%) of the statistical variation in this measurement was accounted for by differences in the use of chemotherapy and hormone
therapy. Treatment data (up to 9 weeks after diagnosis) have been recorded by the Yorkshire Cancer Registry for many years but did not take account of qualitative variations in the agents given, their dosages, and the protocols that may have been in use at the time of data collection. This fact will lead to a bias in underestimating the relation between treatments and outcome. There are likely to have been many suboptimal treatments given over the years in question. In addition, the problems of using a longterm database in which changes in practice occur with time (as well as retirements and new appointments) must not be minimised. The benefits of chemotherapy and hormone therapy may, therefore, have been understated although we can offer no evidence for the magnitude of this effect. The clear relation between these two therapies and outcome is consistent with the many studies that have shown outcome benefits from these interventions. Such evidence arises from clinical trials under controlled conditions and it is important to see that this can be confirmed on a population basis. Another explanation for the significance of the links between use of chemotherapy and hormone therapy, case load, and survival might lie in the value, now widely accepted, of a multidisciplinary approach to the management of breast cancer. This has the objective of ensuring that all relevant treatment methods that have a place in modern management are properly considered and deployed when necessary. Higher rates of these treatments, taken together, are likely to characterise services working efficiently. Thus some of the improved survival associated with greater consultant caseload might be due to those consultants who treat more patients also making more use of the full range of therapeutic options to the benefit of their patients. It is tempting to estimate the impact on survival that might result from the widespread adoption of higher levels of chemotherapy and hormone therapy. Had chemotherapy been used for 40% of patients (instead of the mean regional usage of 9-3%), the calculations in table 7 predict an improvement in 5-year survival of about 4%, and for hormone therapy (already much more widely adopted) the improvement in survival lies between 1 and 1269
1-5%. These percentages approximate to an additional 500 women still alive-a worthwhile gain. There was no relation between rates of use of radiotherapy and survival. This finding might have been anticipated because the primary role of radiotherapy in breast cancer lies in local control. High rates of surgery alone were significantly associated with a poorer survival, an observation lending support to the belief that multiple therapies, correctly deployed, improve survival. Surgeons with higher caseloads but lower use of multiple therapies are exceptions to the general finding of an improved survival for increased caseload. This is an important confirmatory observation to the suggested interpretation of the results for chemotherapy and hormone therapy. The effect of number of patients treated on outcome of breast cancer may not, therefore, as might be argued for some gastrointestinal cancers, be a simple function of the skill of the surgical team, but is rather a function of clinical organisation. The volume effect reflects the ability to bring together the necessary clinical disciplines and expertise across the full therapeutic range. Gillis et al13 considered the issue of multidisciplinary practice to be relevant to the interpretation of their data on survival from ovarian cancer. It has also been used to explain at least part of the difference in survival for Hodgkin’s disease between centres and peripheral hospitals in the USA. 14 The report on cancer servicesl5 and guideline documents on breast cancer" emphasise moves to improve breast cancer services based on the two principles of surgical site specialisation with higher workloads and multidisciplinary working. Pressure groups are urging women to look for services with these characteristics." Many such changes in services are under way with widespread acceptance of their desirability. These changes are consistent with the evidence on caseload and treatment variations described in this paper.
1270
References 1
2
3 4
5
6 7 8 9
Berino F, Sant M, Verdecchia A, Capocaccia R, Hakulinnen T, Esteve J, eds. Survival for cancer patients in Europe. IARC Sci Publ 132
(in press). Sainsbury JRC,
Rider L, Smith A, McAdam WFA. Does it matter where you live? Treatment variation for breast cancer in Yorkshire. Br J Cancer 1995; 71: 1275-78. Stiller CA. Centralisation of treatment and survival rates for cancer. Arch Dis Child 1988; 63: 23-30. Kramer S, Meadows AT, Pastor G, et al. Influence of place of treatment on diagnosis, treatment and survival in three paediatric solid tumours. J Clin Oncol 1984; 2: 917-23. Matthews HR, Powell DJ, McConkey CC. Effects of the results of surgical experience on the results of resection for oesophageal carcinoma. Br J Surgery 1986; 73: 621-23. Cox DR. Regression models and life tables (with discussion). J R Stat Soc 1972; 34: 187-220. Henderson R. Problems and prediction in survival-data analysis. Stat Med 1955; 14: 161-84. Morphet C. Interpretation of small area census data-area 24/1. London: HM Stationery Office, 1992: 62-72. McArdle CS, Hole D. Impact of variability among surgeons on postoperative morbidity and mortality and ultimate survival. BMJ
1991; 302: 150l-05. 10 Stiller CA. Centralised treatment, entry to trials and survival. Br J Cancer 1994; 70: 352-62. 11 Luft HS, Garnick DW, Mark DH, McPhee SJ. Hospital volume, physician volume and patient outcomes. Ann Arbor: Health Administration Press, 1990. 12 Luft HS, Hunt SS, Maerk SC. The volume-outcome relationship: practice makes perfect or selective referral patterns? Health Services Res 1987; 22: 157-82. 13 Gillis CR, Hold DJ, Still RM, et al. Medical audit, cancer registration and survival in ovarian cancer. Lancet 1991; 337: 611-12. 14 Davis S, Dahlberg S, Myers M, et al. Hodgkin’s disease in the United States: a comparison of patient characteristics and survival in the centralised cancer patient data system and the surveillance, epidemiology and end results program. J Natl Cancer Inst 1987; 78: 471-78. 15 Consultative document: a policy framework for commissioning cancer services. London: Department of Health, 1994. 16 The provision of breast services in the UK: the advantages of specialist breast units. Report of a working party of the British Breast Group. Edinburgh: Churchill Livingstone, 1995. 17 The MacMillan directory of breast cancer services in the UK. London: Cancer Relief MacMillan Fund and The King’s Fund. 1995.