lnt. J. Cancer: 52,164-1 65 ( 1992)
Publicationof the InternationalUnion Against Cancer Publicationde I’UntonInternationaleContre le Cancer
0 1997 Wiley-Liss, Inc.
LETTER TO THE EDITOR Dear Sir, Human papilloma virus infection and other risk factors for cervical neoplasia
While the results reported by Morrison et a]. (1991) support an etiologic role for human papilloma virus (HPV) in the development of cervical neoplasia, care is needed in intetpreting these findings. Firstly, there appears to be a critical bias in the selection of cases and controls. While cases were drawn from women referred to a hospital colposcopy clinic, controls were selected from family planning and other gynecologic clinics. The authors do not say how the controls were chosen but these women were likely to be different from the cases in several respects. Certainly they were younger. Since age is associated with severity of cervical intraepithelial neoplasia (CIN), why was no attempt made to match for age? Several studies have found oral contraceptive use to be a positive risk factor for cervical neoplasia ( H a m s et al., 1980; WHO, 1985; Beral et a]., 1988; Vessey et al., 1989) yet controls were taken from family planning clinics where one would expect a greater prevalence ofpill use, as indeed was the case. There was also a statistically significant difference in the percentage of women who were excluded from the analysis because of pregnancy (cases 22%, controls 56%, p = 0.01). This also suggests that the controls came from a different population of women. It might also be expected that data from women attending family planning and other gynecologic clinics could be biased with respect to other established risk factors for cervical neoplasia such as socio-economic status and number ofprevious smears. The I988 Bethesda System for reporting cervicallvaginal cytological diagnoses recommends that cellular changes associated with HPK even without features of “dysplasia” or “CIN”shou1d be included with cases of “low-gradesquamous intraepithelial lesions” (National Cancer Institute Workshop, 1989). i n this study, cases of koilocytotic atypia were included with those of CIN I. While this inclusion was operationally correct, the condition is almost pathognomic of HPV infection (Meisels and Alonso de Ruiz, 1989). Thus, women who were veiy likely to have been infected with HPV were included as cases even though there was no histological evidence of ClN. This would overestimate the prevalence of infection in cases, a methodological problem of studies of HPVand cervical neoplasia addressed by Mutioz et al. (1988). The authors do not state how many cases with koilocytotic atypia there were. It would be of interest to know if the exclusion of these women from the analyses made any difference to the results. It is also of interest that the percentage of cases was similar in each of the CIN groups (CIN I (35%), CIN 11 (32%), CIN 111 (33%)). One might have expected the percentage to decrease as the ClN grade increased. The study evaluated the prevalence of HPV infection using 2 different techniques, Southem-blot hybridization ( S B ) and polymerase chain reaction (PCR).Although more viral types of HPV infection were investigated using the SB technique, a higher proportion of women were identified as having HPV
when PCR was used. For controls in particular, the percentage infected increasedfrom 16% to 39%. It may be that the PCR test has a lower specificity than the SB test, which could explain the lower risk of cervical neoplasia associated with infection when PCR was used ( O R = 10.4) as compared to the risk when SB was used ( O R = 17.9). The role of misclassijication of HPV infection in epidemiologic studies has been recently addressed by Franco ( I 991) who has demonstrated that even low levels of non-diflerential misclassification (mainly lack of specificity) produce a substantial bias in the odds ratio. In addition, in the analysis using the PCR technique, there were no data on 18/83 cases (22%) and 2/61 controls (3%) who had been included in the analysis using the SB technique. This disproportionate loss of cases in the PCR analysis could also have biased the true odds ratio. Another explanation for these differences is the ability of PCR to better identify low-level infection. It is possible, as Morrison et al. suggest, that the PCR-positive but cytologically normal women may have had an inactive latent infection or a viral load insufficient to cause disease. The results in their Table 111, showing higher odds ratios with increased signal intensity, are suggestive of this. The authors suggest a trend toward increasing risk of severity of CINgrade with HPVinfection. Their Table Vshowed this to be statistically significant for both the S B and PCR analyses (p < 0.01). This signijicant trend must have been calculated using the controls as the reference group. This is inappropriate. We have calculated the x2for trend among cases only (Mantel, I963). It was not possible to do this exactly as the CIN grade was not given for the cases who were excluded from the analyses. Since only 2 cases were excluded from the S B analysis we calculated the trend using all cases. For the PCR analysis 65/85 (76.5%) cases were included. W e applied this percentage to the totals in each CIN grade and then, using the data given in Table V, estimated the numbers of women who were +PCR and -PCR. The results are shown in Table I. TABLE I - PREVALENCE OF HPV INFECTION ACCORDING TO CIN GRADE
E$’/ HPV
=
1.08,~ = 0.30
Eii{ HPV
CINIII
22 (73%) 23 (85%) 17 (62%) 8 4 11 30 27 28
+SB -SB
x’ for trend
CIN II
+PCR - PCR
x ? for trend = 1.25, E, = 0.26
‘Koilocytoticatypia.
CIN I1
CINIII
17 (76%) 20 (95%) 18 (85%) 6 1 3 23 21 21
HPV INFECTION AND CERVICAL NEOPLASIA
Thus, there I S no evidence of trend when only the cases are considered. We suggest that the very high atid statistically signijicant odds ratios associated with all the features of HPV infection are biased upwards because of methodological problems, particuladv selection bias. That the results from this study found no increased risk of cenjical neoplasia associated with several of the established risk factors (Tuble II), lends support to this view. Yours sincerely, Jose E L U F - N E Tand ~ ~Margaret I~ BOOTH^
165
'Department of Preventive Medicine, Faculdade de Medicina USP, Sdo Paulo, Brazil; and 'Epidemiological NIonitoring Unit, Department of EpidemioloCy and Population Scietlces, London School of Hygiene and Tropical Medicine, Keppel Street, London W C l E 7 H z UK. March 27, 1992.
?Present address: Department of Epidemiology and Population Sciences, London School of Hygiene and Tropical Medicine, Keppel Street, London WClE 7HT, UK.
REFERENCES
BE.RAL,V., HANNAFOKD. P. and K A Y , C., Oral contraceptive use and malignancies of the genital tract. Results from the Royal College of General Practioners' oral contraception study. Lancer, 11, 1331-1335 (1988). FRANCO, E.L., The sexually transmitted disease model for cervical cancer: incoherent epidemiologic finclings and the role of misclassification of human papillomavirus infection. Epidemiology, 2, 98-106 ( 1 991 ). HARRIS.R.W.C., BRINTON. L.A., COWDELL, R.H., SKEGG,D.C.G., SMITH,P.G., VESS~Y, M.P. and DOLL,R., Characteristics of women with dysplasia or carcinoma in sitir of Ihe cervix uteri. Brit. J. Cancer, 42, 359-369 (lY80). MANTbL. N., Chi-square tests with one degree of freedom: extensions of the Mantel-Haenszel procedure. J. Amer. Stntist. Ass., 58, 690-700 (1963). MEISEI-S, A. and ALONSODE Rurz, P., Human papillomavirus-related changes in the genital tract. In: N. Mufioz, F.X. Bosch and O.M.
Jensen (eds.), Human papillomavirus arid cervical cancer, pp. 67-85, IARC, Lyon (1989). MORRISON, E.A.B., Ho, G.Y.F., VERMUND, S.H., GOLDBERG, G.L., A.S., KELLEY,K.F. and BURK.R.D., Human papillomavirus KADISH, infection and other risk factors for cervical neoplasia: a case-control study. In[. J. Cancer, 49, 6-13 (1991). Mukoz, N., BOSCH,F.X. and KALDOR,J.M., Does human papillomavirus cause cervical cancer? The state of the epidemiological evidence. Brit. J. Cancer, 57,l-5 (1988). NATIONAL CANCER INSTITUTE WORKSHOP. The 1988 Bethesda System for reporting cervicalivaginal cytological diagnoses. J. Amer. med. Ass., 262,931-934 (1989). VESSEY, M.P., VILLARD-MACKINTOSH, L., MCPHERSON, K. and YEATES, D., Mortality among oral contraceptive users: 20 year follow-up of women in a cohort study. Brit. med. J., 299,1487-1491 (1980). WHO COLLABORATIVE STUDY OF NEOPLASIAAND STEROID CONTRACEPTIVES. Invasive cervical cancer and combined oral contraceptives. Brit. med. J.. 290,961-965 (1985).
Publicationof the InternationalUnion Against Cancer Publicationde I’UntonInternationaleContre le Cancer
0 1997 Wiley-Liss, Inc.
LETTER TO THE EDITOR Dear Sir, Human papilloma virus infection and other risk factors for cervical neoplasia
While the results reported by Morrison et a]. (1991) support an etiologic role for human papilloma virus (HPV) in the development of cervical neoplasia, care is needed in intetpreting these findings. Firstly, there appears to be a critical bias in the selection of cases and controls. While cases were drawn from women referred to a hospital colposcopy clinic, controls were selected from family planning and other gynecologic clinics. The authors do not say how the controls were chosen but these women were likely to be different from the cases in several respects. Certainly they were younger. Since age is associated with severity of cervical intraepithelial neoplasia (CIN), why was no attempt made to match for age? Several studies have found oral contraceptive use to be a positive risk factor for cervical neoplasia ( H a m s et al., 1980; WHO, 1985; Beral et a]., 1988; Vessey et al., 1989) yet controls were taken from family planning clinics where one would expect a greater prevalence ofpill use, as indeed was the case. There was also a statistically significant difference in the percentage of women who were excluded from the analysis because of pregnancy (cases 22%, controls 56%, p = 0.01). This also suggests that the controls came from a different population of women. It might also be expected that data from women attending family planning and other gynecologic clinics could be biased with respect to other established risk factors for cervical neoplasia such as socio-economic status and number ofprevious smears. The I988 Bethesda System for reporting cervicallvaginal cytological diagnoses recommends that cellular changes associated with HPK even without features of “dysplasia” or “CIN”shou1d be included with cases of “low-gradesquamous intraepithelial lesions” (National Cancer Institute Workshop, 1989). i n this study, cases of koilocytotic atypia were included with those of CIN I. While this inclusion was operationally correct, the condition is almost pathognomic of HPV infection (Meisels and Alonso de Ruiz, 1989). Thus, women who were veiy likely to have been infected with HPV were included as cases even though there was no histological evidence of ClN. This would overestimate the prevalence of infection in cases, a methodological problem of studies of HPVand cervical neoplasia addressed by Mutioz et al. (1988). The authors do not state how many cases with koilocytotic atypia there were. It would be of interest to know if the exclusion of these women from the analyses made any difference to the results. It is also of interest that the percentage of cases was similar in each of the CIN groups (CIN I (35%), CIN 11 (32%), CIN 111 (33%)). One might have expected the percentage to decrease as the ClN grade increased. The study evaluated the prevalence of HPV infection using 2 different techniques, Southem-blot hybridization ( S B ) and polymerase chain reaction (PCR).Although more viral types of HPV infection were investigated using the SB technique, a higher proportion of women were identified as having HPV
when PCR was used. For controls in particular, the percentage infected increasedfrom 16% to 39%. It may be that the PCR test has a lower specificity than the SB test, which could explain the lower risk of cervical neoplasia associated with infection when PCR was used ( O R = 10.4) as compared to the risk when SB was used ( O R = 17.9). The role of misclassijication of HPV infection in epidemiologic studies has been recently addressed by Franco ( I 991) who has demonstrated that even low levels of non-diflerential misclassification (mainly lack of specificity) produce a substantial bias in the odds ratio. In addition, in the analysis using the PCR technique, there were no data on 18/83 cases (22%) and 2/61 controls (3%) who had been included in the analysis using the SB technique. This disproportionate loss of cases in the PCR analysis could also have biased the true odds ratio. Another explanation for these differences is the ability of PCR to better identify low-level infection. It is possible, as Morrison et al. suggest, that the PCR-positive but cytologically normal women may have had an inactive latent infection or a viral load insufficient to cause disease. The results in their Table 111, showing higher odds ratios with increased signal intensity, are suggestive of this. The authors suggest a trend toward increasing risk of severity of CINgrade with HPVinfection. Their Table Vshowed this to be statistically significant for both the S B and PCR analyses (p < 0.01). This signijicant trend must have been calculated using the controls as the reference group. This is inappropriate. We have calculated the x2for trend among cases only (Mantel, I963). It was not possible to do this exactly as the CIN grade was not given for the cases who were excluded from the analyses. Since only 2 cases were excluded from the S B analysis we calculated the trend using all cases. For the PCR analysis 65/85 (76.5%) cases were included. W e applied this percentage to the totals in each CIN grade and then, using the data given in Table V, estimated the numbers of women who were +PCR and -PCR. The results are shown in Table I. TABLE I - PREVALENCE OF HPV INFECTION ACCORDING TO CIN GRADE
E$’/ HPV
=
1.08,~ = 0.30
Eii{ HPV
CINIII
22 (73%) 23 (85%) 17 (62%) 8 4 11 30 27 28
+SB -SB
x’ for trend
CIN II
+PCR - PCR
x ? for trend = 1.25, E, = 0.26
‘Koilocytoticatypia.
CIN I1
CINIII
17 (76%) 20 (95%) 18 (85%) 6 1 3 23 21 21
HPV INFECTION AND CERVICAL NEOPLASIA
Thus, there I S no evidence of trend when only the cases are considered. We suggest that the very high atid statistically signijicant odds ratios associated with all the features of HPV infection are biased upwards because of methodological problems, particuladv selection bias. That the results from this study found no increased risk of cenjical neoplasia associated with several of the established risk factors (Tuble II), lends support to this view. Yours sincerely, Jose E L U F - N E Tand ~ ~Margaret I~ BOOTH^
165
'Department of Preventive Medicine, Faculdade de Medicina USP, Sdo Paulo, Brazil; and 'Epidemiological NIonitoring Unit, Department of EpidemioloCy and Population Scietlces, London School of Hygiene and Tropical Medicine, Keppel Street, London W C l E 7 H z UK. March 27, 1992.
?Present address: Department of Epidemiology and Population Sciences, London School of Hygiene and Tropical Medicine, Keppel Street, London WClE 7HT, UK.
REFERENCES
BE.RAL,V., HANNAFOKD. P. and K A Y , C., Oral contraceptive use and malignancies of the genital tract. Results from the Royal College of General Practioners' oral contraception study. Lancer, 11, 1331-1335 (1988). FRANCO, E.L., The sexually transmitted disease model for cervical cancer: incoherent epidemiologic finclings and the role of misclassification of human papillomavirus infection. Epidemiology, 2, 98-106 ( 1 991 ). HARRIS.R.W.C., BRINTON. L.A., COWDELL, R.H., SKEGG,D.C.G., SMITH,P.G., VESS~Y, M.P. and DOLL,R., Characteristics of women with dysplasia or carcinoma in sitir of Ihe cervix uteri. Brit. J. Cancer, 42, 359-369 (lY80). MANTbL. N., Chi-square tests with one degree of freedom: extensions of the Mantel-Haenszel procedure. J. Amer. Stntist. Ass., 58, 690-700 (1963). MEISEI-S, A. and ALONSODE Rurz, P., Human papillomavirus-related changes in the genital tract. In: N. Mufioz, F.X. Bosch and O.M.
Jensen (eds.), Human papillomavirus arid cervical cancer, pp. 67-85, IARC, Lyon (1989). MORRISON, E.A.B., Ho, G.Y.F., VERMUND, S.H., GOLDBERG, G.L., A.S., KELLEY,K.F. and BURK.R.D., Human papillomavirus KADISH, infection and other risk factors for cervical neoplasia: a case-control study. In[. J. Cancer, 49, 6-13 (1991). Mukoz, N., BOSCH,F.X. and KALDOR,J.M., Does human papillomavirus cause cervical cancer? The state of the epidemiological evidence. Brit. J. Cancer, 57,l-5 (1988). NATIONAL CANCER INSTITUTE WORKSHOP. The 1988 Bethesda System for reporting cervicalivaginal cytological diagnoses. J. Amer. med. Ass., 262,931-934 (1989). VESSEY, M.P., VILLARD-MACKINTOSH, L., MCPHERSON, K. and YEATES, D., Mortality among oral contraceptive users: 20 year follow-up of women in a cohort study. Brit. med. J., 299,1487-1491 (1980). WHO COLLABORATIVE STUDY OF NEOPLASIAAND STEROID CONTRACEPTIVES. Invasive cervical cancer and combined oral contraceptives. Brit. med. J.. 290,961-965 (1985).
