be either higher or lower than 1, i.e., having serological mark-to the exposure of interest, constitute a valid and well-accepted comparison group, which has been used in countless ers of infection by hepatitis B and C viruses could be either ''protective'' or ''predisposing'' factors for aplastic anemia (or case-control studies. Specifically with regard to the present study, the selection of controls whose admission was obliga-interpreted as surrogates for other unknown factors correlated to them).
tory because of abdominal emergencies (e.g., appendicitis), acute infections (e.g., pneumonia), or trauma, is unlikely A possible way to assess whether this criticism is relevant is by re-analyzing the data using conditional logistic regression, to be related to previous hepatitis. Additional controls with elective admissions for procedures such as cosmetic surgery restricting the analysis to the 123 cases and 129 matched controls examined for serological markers of hepatitis A, B, were carefully selected so that their reason for admission would be independent of risk factors such as hepatitis. and C. This should provide a more valid (although less precise) estimate of the association between seropositivity to To maximize statistical efficiency, the matching factors, age, sex, and region, were individually controlled by uncon-these viruses and aplastic anemia, provided that the subjects included in this analysis are a random sample of the original ditional logistic regression; the relative risk estimates for hepatitis A, B, and C reported in the article were 2.9 (95% confi-study population (matched cases and controls). Dr. Issaragrisil et al. are in position to state whether this assumption dence interval, 1.2-6.7), 1.2 (0.7-2.0), and 1.2 (0.5-3.3), respectively. The conditional logistic model suggested by Dr. holds.
Nishioka yielded corresponding estimates of 3.9 (1.1-14), 1.2 (0.5-2.8), and 1.3 (0.2-6.7). Given the considerably wider SE Β΄RGIO DE A. NISHIOKA, M.D., M.SC. confidence intervals that result from the latter procedure, Centro de Cie Λncias Biome Β΄dicas, the two sets of estimates are statistically compatible. In addi-Universidade Federal de Uberla Λndia tion, the conditional estimates do not alter any of the conclu-Uberla Λndia, Brazil sions. HEPATOLOGY December 1997 (72.3%) of our patients had single tumors less than 4 cm in VIJAYAN BALAN, M.B.B.S. JORGE RAKELA, M.D. diameter. Additionally, the comparison of our model with the TNM classification system, the model most commonly Department of Biomedical Infomatics used 4,5 for prediction of cancer patients' outcomes, showed that out of 65 patients assigned to our non-recurrence group, VLADIMIR SUBBOTIN, M.D., PH.D. 22 (33.8%), 21 (32.3%), 20 (30.8%), and 2 (3.1%) had TNM Department of Pathology stages I, II, III, and IV-A, respectively, while among 44 pa-University of Pittsburgh tients predicted by our system to have certain recurrence, 2 Pittsburgh, PA (4.5%), 39 (88.6%), and 3 (6.8%) belonged to stages III, IV-A, and IV-B, respectively. These numbers suggest that our E. P. POPECHITELEV, PH.D. model could correctly identify an additional 27.7% of pa-Department of Biological Systems tients over the model based on tumor number and tumor St. Petersburg Electrotechnical University size, 3 and an additional 33.9% over the selection of transplant St. Petersburg, Russia candidacy based on TNM stage (i.e., patients in either stages I or II). It should be noted that these percentages do not take REFERENCES into account the additional patients who, based on model 1. Ringe B. Can the prognosis of liver transplantation for hepatocellular predictions, would not have developed recurrent hepatocelcarcinoma be predicted. HEPATOLOGY 1996;26(2):444-450. lular carcinoma if they had not died within the first 3 years 2. Marsh WJ, Dvorchik I, Subotin M, Balan VJ, Popechitelev EP, Subbotin after surgery from other causes. In our opinion, such an V, Casavilla A, et al. The prediction of risk of recurrence and time to recurrence of hepatocellular carcinoma after orthotopic liver transplanta-advantage in patient selection makes the additional sophistition: a pilot study. HEPATOLOGY 1997;26(2):507-508. cation of statistical analysis worthwhile.
- McPeake JR, O'Grady JG, Zaman S, Portmann B, Wight DGD, Tan KC, Dr. Ringe's justifiable concerns about various impondera-Calne RY et al. Liver transplantation for primary hepatocellular carcibles that prevent unambiguous prognostication of the postnoma: tumor size and number determine outcome. J Hepatol 1993;18: transplant course for an individual suffering from hepatocel-226-234. lular carcinoma are taken into account by the indeterminate 4. Ringe B, Wittekind C, Bechstein WO, Bunzendahl H, Pichlmayr R. The role of liver transplantation in hepatobiliary malignancy. A retrospective category of our predictions. The occurrence of diverging outanalysis of 95 patients with particular regard to tumor stage and recurcomes for the same combinations of risk factors is caused by