778
patients referred for a gynaecological opinion. But are we right not to be surprised at the higher figure? In-situ squamous-cell carcinoma of the cervix is reportedly asymptomatic. Pregnancy can produce very atypical cellular patterns, some of which return to normal after delivery. All the same the rate for histologically proven malignancy in patients attending the antenatal clinic (10 per 1000) seems high. The last group in table n are those smears taken by general practitioners, with a pick-up rate of 8 per 1000, which is all,
not
well
women
but
twice the national average. This supports our view that G.P.S should take more smears. With their knowledge of their patients’ medical and social histories they can collect smears from those at greatest risk. Colposcopy is a useful intermediate step between an abnormal smear and cone biopsy. Previously all patients with an abnormal smear required a cone biopsy, with all the associated complications. Before colposcopy was introduced cytopathologists knew that an abnormal report would result in a cone biopsy or even in hysterectomy without biopsy, but now that conservative management of abnormal smears by colposcopy is available conservative treatment of the woman with an abnormal smear, particularly during reproductive life, is not only practicable but also desirable. Fig. 1 demonstrates that the number of cytologically abnormal smears (and proven histologically malignant cases) reaches a peak in the age-group 26-30. In 1976 Livingston and Joel6 noted an 800% increase in the number of suspicious Papanicolaou smears in the younger age-group they routinely examined. The clinical histories obtained pointed to two factors, promiscuity and oral contraception. In 115 cases of "surface dysplasia" 8 were attributed to promiscuity, 28 to birth-control pills, and in 67 cases both factors were implicated. World wide there seems to be an increase in the numbers of young women who have abnormal smears. The Walton report4 refers to a peak incidence of dysplasia and carcinoma-in-situ in women in their twenties. There is a long interval between this point and the onset of invasive cancer. Yule7 noted a small decrease in the number of deaths from carcinoma of the cervix in England and Wales, with significant improvement in the age-group 35-54 years; but the number of deaths in women under 35 years of age had doubled. Macgregor and Teperfeel that routine screening of young girls is of doubtful value; early pregnancy, they suggest, is the key factor in the genesis of cervical cancer, and oral contraception, by reducing the number of early pregnancies, could reduce the incidence of the disease. Several causative factors have been put forward for cervical cancer (e.g., smegma, trichomonads, a virus, poor hygiene, and the basic proteins of the sperm head9). The modern concept of the genesis of cervical malignancylO presumes that coitus-associated material acts as a carcinogen, with the cells of "active metaplasia" as target cells, producing early cellular abnormalities and a "suspicious" smear. Promiscuous young women have an increased incidence of metaplasia which appears to be initiated and maintained by coitus. Leppaluoto" has drawn attention to the fact that women who use birthcontrol pills have an increased incidence of cervical ectopy, and .in ectopy exposure of the endocervical col-
cells to the acid vaginal secretions results in the cervical metaplasia (i.e., target cells ready for a chance hit by a coitus-associated carcinogen). Thus both groups of women would present an increased frequency of "sus-
umnar
picious" Papanicolaou smears. Lepp;xluoto points out that the ideal contraceptive, from an oncogenic viewpoint, would be one that restores the cervical epithelium to normal health, thus helping to prevent cervical malignancy. This approach is at variance with Macgregor’s notion that oral contraceptives, by reducing the number of pregnancies, may reduce the incidence of the disease. Until the aetiology of cervical cancer is established we believe that all young women should be encouraged to have cervical smears. This view is compatible with the concept of limiting cytological screening to high-risk groups, for many young gory.
women now come
into this
We thank the gynaecologists and their staff for their Mrs Jean Wiles for typing the manuscript. Requests for reprints should be addressed to D.H.M.
cate-
cooperation and
REFERENCES 1.
Christopherson, W. M., Mendez, W. M., Ahuja, E. M., Lundin, F. E., Parker, J. E. Cancer, 1970, 26, 808. 2. Macgregor, J. E., Teper, S. Lancet, 1974, i, 1221. 3. Timonen, S., Nieminen, V., Kauraniemi, T. ibid. 1974, i, 401. 4. Walton Report. Can. med. Ass. J. 1976, 114, 971. 5. Melcher, D. H., Linehan, J. J. Lancet, 1974, i, 1221. 6. Livingston, G. A., Joel, R. V. Acta cytol. 1976, 20, 4. 7. Yule, R. Lancet, 1978, i, 1030. 8. Macgregor, J. E., Teper, S. ibid. 1978, i, 1029. 9. Reid, B. L., French, P. W., Singer, A., Hagan, B. E., Coppelson, M. ibid. 1978, ii, 60. 10. Coppelson, M., Pixley, E. C., Reid, B. L., Singer, A. Scientific exhibit no. 23 in Sixth World Congress of Gyæcology & Obstetrics, New York, April, 1970. 11.
Leppäluoto,
P. Acta
cycol. 1977, 21,
Hospital
182.
Practice
SURGEON-RELATED VARIABLES AND THE CLINICAL TRIAL L. P. FIELDING
SARAH STEWART-BROWN H. A. F. DUDLEY
Academic Surgical Unit, St. Mary’s Hospital Medical School, London W2 1PG
trial two preconditions for standardisation of all methods of treatment except the aspect being studied and randomisation so that variation in the population studied is approximately equally distributed between the control and treatment groups. In a multicentre trial, standardisation can be difficult because of the different treatment styles of the participating doctors. In such trials randomisation becomes the chief method of avoiding bias, and to counter the increased variation it is customary to recommend that the trial size be larger than that for a similar study done in a single institution. Random variance ought to decrease as numbers increase, but in practice this does not always follow. Data from our study in colorectal cancer illustrate a concept which we feel has wider IN
a
prospective clinical
success are
application. Calculation of the number of patients required is an imporpart of trial design. One aspect of this calculation is the
tant
779
.
identification of all significant predictive (prognostic) variables-factors that, from biological reasoning or past experience, are known to influence or be determinant of outcome. They can be divided into two groups: patient and treatment (surgeon) related variables. For colorectal cancer there are eight main patient-related predictive variables: age (above or below 70), presentation (elective or emergency), presence or absence of other severe disease such as cardiorespiratory illness or diabetes, use of drugs (steroids/anticoagulants), local spread (Dukes A or B), lymph-node involvement, distant spread, and differentiation (moderate or poor). The number of different patient types in the population is xn, where x is the number of states for each variable (here 2, such as present or absent) and n is the number of variables (8). Thus theoretically there are 28 or 512 different patient types in colorectal cancer. Even if only 150 of these possible types are common, some 6000 patients would be needed to achieve just 20 of each type in a two-armed study. We have not read a single report of a randomised trial in which any attempt has been made to demonstrate matching of the groups for surgeon. To investigate this we have analysed the results of initial treatment in 932 patients in a multicentre trial in primary colorectal adenocarcinoma. Sepsis was a common cause of both morbidity and mortality, and postoperative sepsis is known to be influenced by bowel preparation, perioperative antibiotics, and surgical tech-
nique (in that a reduction in anastomotic leakage should reduce mortality and must reduce local sepsis). The variation in these three factors is shown in figs. 1 and 2. The wide range of mortality (fig. 3) suggests that outcome is to a considerable degree surgeon-dependent. The "same" operation may have a different outcome in the hands of different surgeons. Thus standardisation of treatment cannot be achieved and a multicentre trial becomes a study of a variety of treatments in a variety of patients. Consider the practical implications. In colorectal cancer, randomised trials are being widely used to assess the efficacy of adjuvant therapy after resection--e.g., surgery plus adjuvant regimen A is compared with surgery plus adjuvant regimen B (which may be placebo). Since the effect of surgery is not a constant in this equation no conclusions can be drawn about the comparative efficacy of regimens A and B. One way of overcoming this problem would be to list the predictive variables related to surgeon and to treat these as we treated those related to the patient, using, for example, stratification to obtain similarity in the study groups or, where stratitification is not possible, demonstrating when the results are reported that randomisation has succeeded in equally distributing the variables. We have drawn up a list of five surgeonrelated predictive variables, giving 25 or 64 different surgeon types. These are: experience (trainee or specialist) surgical technique (conventional dissection or "no-touch" dissection), use of deep-vein thrombosis prophylaxis, presence or absence of postoperative systemic sepsis, and postoperative mortality. However, surgeon-related predictive variables summate with patient-related ones, so the number of postoperative patient types becomes 28+5 or 8192. The number of patients required to achieve any reasonable number of each postoperative patient type in a two-armed study thus becomes prohibitively
large.
study of large-bowel cancer has not been running long enough for us to assess the effect of surgeon on long-term survival. However, we probably will be able to do this because the population of patients on whom survival can be measured is partly defined both in numbers of patients and in their immunological status by the contributing surgeon’s operative mortality and sepsis-rate. Our multicentre
Fig. 1-Types of bowel preparation and, within each type, use of systemic perioperative antibiotics. Mech.=mechanical only; Antim.=antimicrobial agent only; E.D.=elemental diet; others=combination of these types.
Any one surgeon has only a limited number of patients under his care who can be entered into a trial (on average about 20 a year). To decrease random variance in the patient population increased numbers are required, but the numbers can only be increased by increasing either the number of surgeons or the duration of the trial. However, increasing the number of surgeons may increase variance; and increasing the length of the trial may have a similar effect because of the difficulty of keeping all aspects of treatment constant during a
prolonged study. /o
of clinical anastomotic leaks in operations done by surgeons contributing 10 or more patients.
Fig. 2-Frequency
/0
-
Fig. 3-Postoperative mortality for surgeons contributing 10 or more patients.
after elective tumour resection’
The nature and number of patient-related predictive variables are fixed by the patient and his disorder, but surgeon-related predictive variables can be changed. One strategy might be to do a prospective study to identify surgeon-related predictive variables and then, by discussion with and with the cooperation of those who wish to take part in the main trial, to reduce or eliminate these variables. Unfortunately, the need for this type of preliminary study is seldom recognised by those wishing to conduct a trial or by grant-giving bodies. The presentation of the data for this paper would not have been possible without the cooperation of the 62 surgeons in twenty-three hospitals involved in the multicentre large-bowel cancer study to whom we are grateful. We thank Prof. M. J. R. Healy for his comments and Downs Surgical Ltd and the Heckford, Norton & Co. bequest for financial support. Requests for reprints should be addressed to H.A.F.D.
patients referred for a gynaecological opinion. But are we right not to be surprised at the higher figure? In-situ squamous-cell carcinoma of the cervix is reportedly asymptomatic. Pregnancy can produce very atypical cellular patterns, some of which return to normal after delivery. All the same the rate for histologically proven malignancy in patients attending the antenatal clinic (10 per 1000) seems high. The last group in table n are those smears taken by general practitioners, with a pick-up rate of 8 per 1000, which is all,
not
well
women
but
twice the national average. This supports our view that G.P.S should take more smears. With their knowledge of their patients’ medical and social histories they can collect smears from those at greatest risk. Colposcopy is a useful intermediate step between an abnormal smear and cone biopsy. Previously all patients with an abnormal smear required a cone biopsy, with all the associated complications. Before colposcopy was introduced cytopathologists knew that an abnormal report would result in a cone biopsy or even in hysterectomy without biopsy, but now that conservative management of abnormal smears by colposcopy is available conservative treatment of the woman with an abnormal smear, particularly during reproductive life, is not only practicable but also desirable. Fig. 1 demonstrates that the number of cytologically abnormal smears (and proven histologically malignant cases) reaches a peak in the age-group 26-30. In 1976 Livingston and Joel6 noted an 800% increase in the number of suspicious Papanicolaou smears in the younger age-group they routinely examined. The clinical histories obtained pointed to two factors, promiscuity and oral contraception. In 115 cases of "surface dysplasia" 8 were attributed to promiscuity, 28 to birth-control pills, and in 67 cases both factors were implicated. World wide there seems to be an increase in the numbers of young women who have abnormal smears. The Walton report4 refers to a peak incidence of dysplasia and carcinoma-in-situ in women in their twenties. There is a long interval between this point and the onset of invasive cancer. Yule7 noted a small decrease in the number of deaths from carcinoma of the cervix in England and Wales, with significant improvement in the age-group 35-54 years; but the number of deaths in women under 35 years of age had doubled. Macgregor and Teperfeel that routine screening of young girls is of doubtful value; early pregnancy, they suggest, is the key factor in the genesis of cervical cancer, and oral contraception, by reducing the number of early pregnancies, could reduce the incidence of the disease. Several causative factors have been put forward for cervical cancer (e.g., smegma, trichomonads, a virus, poor hygiene, and the basic proteins of the sperm head9). The modern concept of the genesis of cervical malignancylO presumes that coitus-associated material acts as a carcinogen, with the cells of "active metaplasia" as target cells, producing early cellular abnormalities and a "suspicious" smear. Promiscuous young women have an increased incidence of metaplasia which appears to be initiated and maintained by coitus. Leppaluoto" has drawn attention to the fact that women who use birthcontrol pills have an increased incidence of cervical ectopy, and .in ectopy exposure of the endocervical col-
cells to the acid vaginal secretions results in the cervical metaplasia (i.e., target cells ready for a chance hit by a coitus-associated carcinogen). Thus both groups of women would present an increased frequency of "sus-
umnar
picious" Papanicolaou smears. Lepp;xluoto points out that the ideal contraceptive, from an oncogenic viewpoint, would be one that restores the cervical epithelium to normal health, thus helping to prevent cervical malignancy. This approach is at variance with Macgregor’s notion that oral contraceptives, by reducing the number of pregnancies, may reduce the incidence of the disease. Until the aetiology of cervical cancer is established we believe that all young women should be encouraged to have cervical smears. This view is compatible with the concept of limiting cytological screening to high-risk groups, for many young gory.
women now come
into this
We thank the gynaecologists and their staff for their Mrs Jean Wiles for typing the manuscript. Requests for reprints should be addressed to D.H.M.
cate-
cooperation and
REFERENCES 1.
Christopherson, W. M., Mendez, W. M., Ahuja, E. M., Lundin, F. E., Parker, J. E. Cancer, 1970, 26, 808. 2. Macgregor, J. E., Teper, S. Lancet, 1974, i, 1221. 3. Timonen, S., Nieminen, V., Kauraniemi, T. ibid. 1974, i, 401. 4. Walton Report. Can. med. Ass. J. 1976, 114, 971. 5. Melcher, D. H., Linehan, J. J. Lancet, 1974, i, 1221. 6. Livingston, G. A., Joel, R. V. Acta cytol. 1976, 20, 4. 7. Yule, R. Lancet, 1978, i, 1030. 8. Macgregor, J. E., Teper, S. ibid. 1978, i, 1029. 9. Reid, B. L., French, P. W., Singer, A., Hagan, B. E., Coppelson, M. ibid. 1978, ii, 60. 10. Coppelson, M., Pixley, E. C., Reid, B. L., Singer, A. Scientific exhibit no. 23 in Sixth World Congress of Gyæcology & Obstetrics, New York, April, 1970. 11.
Leppäluoto,
P. Acta
cycol. 1977, 21,
Hospital
182.
Practice
SURGEON-RELATED VARIABLES AND THE CLINICAL TRIAL L. P. FIELDING
SARAH STEWART-BROWN H. A. F. DUDLEY
Academic Surgical Unit, St. Mary’s Hospital Medical School, London W2 1PG
trial two preconditions for standardisation of all methods of treatment except the aspect being studied and randomisation so that variation in the population studied is approximately equally distributed between the control and treatment groups. In a multicentre trial, standardisation can be difficult because of the different treatment styles of the participating doctors. In such trials randomisation becomes the chief method of avoiding bias, and to counter the increased variation it is customary to recommend that the trial size be larger than that for a similar study done in a single institution. Random variance ought to decrease as numbers increase, but in practice this does not always follow. Data from our study in colorectal cancer illustrate a concept which we feel has wider IN
a
prospective clinical
success are
application. Calculation of the number of patients required is an imporpart of trial design. One aspect of this calculation is the
tant
779
.
identification of all significant predictive (prognostic) variables-factors that, from biological reasoning or past experience, are known to influence or be determinant of outcome. They can be divided into two groups: patient and treatment (surgeon) related variables. For colorectal cancer there are eight main patient-related predictive variables: age (above or below 70), presentation (elective or emergency), presence or absence of other severe disease such as cardiorespiratory illness or diabetes, use of drugs (steroids/anticoagulants), local spread (Dukes A or B), lymph-node involvement, distant spread, and differentiation (moderate or poor). The number of different patient types in the population is xn, where x is the number of states for each variable (here 2, such as present or absent) and n is the number of variables (8). Thus theoretically there are 28 or 512 different patient types in colorectal cancer. Even if only 150 of these possible types are common, some 6000 patients would be needed to achieve just 20 of each type in a two-armed study. We have not read a single report of a randomised trial in which any attempt has been made to demonstrate matching of the groups for surgeon. To investigate this we have analysed the results of initial treatment in 932 patients in a multicentre trial in primary colorectal adenocarcinoma. Sepsis was a common cause of both morbidity and mortality, and postoperative sepsis is known to be influenced by bowel preparation, perioperative antibiotics, and surgical tech-
nique (in that a reduction in anastomotic leakage should reduce mortality and must reduce local sepsis). The variation in these three factors is shown in figs. 1 and 2. The wide range of mortality (fig. 3) suggests that outcome is to a considerable degree surgeon-dependent. The "same" operation may have a different outcome in the hands of different surgeons. Thus standardisation of treatment cannot be achieved and a multicentre trial becomes a study of a variety of treatments in a variety of patients. Consider the practical implications. In colorectal cancer, randomised trials are being widely used to assess the efficacy of adjuvant therapy after resection--e.g., surgery plus adjuvant regimen A is compared with surgery plus adjuvant regimen B (which may be placebo). Since the effect of surgery is not a constant in this equation no conclusions can be drawn about the comparative efficacy of regimens A and B. One way of overcoming this problem would be to list the predictive variables related to surgeon and to treat these as we treated those related to the patient, using, for example, stratification to obtain similarity in the study groups or, where stratitification is not possible, demonstrating when the results are reported that randomisation has succeeded in equally distributing the variables. We have drawn up a list of five surgeonrelated predictive variables, giving 25 or 64 different surgeon types. These are: experience (trainee or specialist) surgical technique (conventional dissection or "no-touch" dissection), use of deep-vein thrombosis prophylaxis, presence or absence of postoperative systemic sepsis, and postoperative mortality. However, surgeon-related predictive variables summate with patient-related ones, so the number of postoperative patient types becomes 28+5 or 8192. The number of patients required to achieve any reasonable number of each postoperative patient type in a two-armed study thus becomes prohibitively
large.
study of large-bowel cancer has not been running long enough for us to assess the effect of surgeon on long-term survival. However, we probably will be able to do this because the population of patients on whom survival can be measured is partly defined both in numbers of patients and in their immunological status by the contributing surgeon’s operative mortality and sepsis-rate. Our multicentre
Fig. 1-Types of bowel preparation and, within each type, use of systemic perioperative antibiotics. Mech.=mechanical only; Antim.=antimicrobial agent only; E.D.=elemental diet; others=combination of these types.
Any one surgeon has only a limited number of patients under his care who can be entered into a trial (on average about 20 a year). To decrease random variance in the patient population increased numbers are required, but the numbers can only be increased by increasing either the number of surgeons or the duration of the trial. However, increasing the number of surgeons may increase variance; and increasing the length of the trial may have a similar effect because of the difficulty of keeping all aspects of treatment constant during a
prolonged study. /o
of clinical anastomotic leaks in operations done by surgeons contributing 10 or more patients.
Fig. 2-Frequency
/0
-
Fig. 3-Postoperative mortality for surgeons contributing 10 or more patients.
after elective tumour resection’
The nature and number of patient-related predictive variables are fixed by the patient and his disorder, but surgeon-related predictive variables can be changed. One strategy might be to do a prospective study to identify surgeon-related predictive variables and then, by discussion with and with the cooperation of those who wish to take part in the main trial, to reduce or eliminate these variables. Unfortunately, the need for this type of preliminary study is seldom recognised by those wishing to conduct a trial or by grant-giving bodies. The presentation of the data for this paper would not have been possible without the cooperation of the 62 surgeons in twenty-three hospitals involved in the multicentre large-bowel cancer study to whom we are grateful. We thank Prof. M. J. R. Healy for his comments and Downs Surgical Ltd and the Heckford, Norton & Co. bequest for financial support. Requests for reprints should be addressed to H.A.F.D.
