HEPATOLOGY Ekewhere HAROLD 0. CONN, EDITOR Yale University School of Medicine Veterans Administration Medical Center West Haven. Connecticut 06516
PARENTERAL NUTRITION WITH BRANCHED-CHAINAMINO ACIDS IN HEPATIC ENCEPHALOPATHY:META ANALYSIS
The report by Naylor et al. not only exemplifies the popularity and attractiveness of the approach, but also underscores some of the problematic issues inherent in Naylor CD, O’Rourke K, Detsky AS, Baker JP. any meta-analysis of clinical trials. This commentary Parenteral nutrition with branched-chain amino acids focuses on one such issue relating to the problem of in hepatic encephalopathy: a meta-analysis. Gastroen- appropriate data abstraction. To evaluate the effect of treatment on mortality, terology 1989;97:1033-1042. Naylor et al. consider four trials (1-4) for “primary” analysis. Table 1 lists the relevant data for these trials EDITOR’S ABSTRACT from the report. This table also lists the corresponding The authors undertook a meta analysis of nine data of a clinical critique of these same trials published randomized clinical trials (RCTs) in which modified by Eriksson and Conn at about the same time (5). Data amino acid solutions that contained supplemental for only one trial (H3) are the same in the two reports. amounts of branched-chain amino acids (BCAA) were Data for the other three trials do not match because of compared with other forms of therapy of hepatic discrepancies in the number of patients studied, the encephalopathy (HE). When five RCTs were pooled, number who survive, or both. The reasons for the analysis showed highly significant improvement in the differences vary: for H2, the “primary” analysis of degree of HE (p < 0.001). Various interpretations of Naylor et al. includes mortality information after the the data gave similar results. The mortality data 48-hour crossover point, whereas Eriksson et al. condiffered so widely in these RCTs that pooling of the sider only the pre-crossover results. For H4, Naylor et al. data could not be meaningfully performed. In two RCTs BCAA therapy resulted in increased mortality, use the data from a preliminary symposium report, whereas in two others BCAA treatment was asso- whereas Eriksson et al. use the data from a subsequent ciated with increased survival. An aggregate relative report ( 6 ) that includes additional patients. For H6, risk reduction of 0.59 was calculated with 95% con- Naylor et al. exclude patients who drop out after fidence intervals of 0.23 to 0.80 (p < 0.002). When the randomization from their “primary” analysis, whereas data from one additional study that had not yet been Eriksson et al. consider all patients who enter the study. published and that showed BCAA therapy to be beneUsing the methods of DerSimonian and Laird (7) to ficial was added to the analysis, the pooled relative evaluate homogeneity of treatment effect, we find that reduction was even greater (i.e., 0.82 [p < 0.00011). The the data from Naylor et al. indicate a large degree mortality in these studies was significantly reduced of heterogeneity (p < 0.021, whereas the data from regardless of the type of statistical analysis performed (p = 0.023). Nevertheless, because of the short du- Eriksson et al. strongly imply homogeneity. These ration of follow-up in these studies and the striking findings hold both when treatment effect is the difdifferences in the effects of BCAA on mortality, the ference in proportions and when it is the log odds ratio. In light of the heterogeneity in the results of authors recommended that additional RCTs with protreatment in the four studies, Naylor et al. sought longed follow-up be performed. explanations for the disparities. Since the trials varied widely in the duration of therapy and in the intake of COMMENTS calories, they divided the trials into two subsets of Meta-analysis is a general term to describe the similar length of follow-up and caloric intake with two statistical analysis of a group of results of individual trials each (Table 1).Within each subset, the homogeinvestigations for the purpose of integrating the find- neity of results of therapy was more homogeneous. This ings. Since they allow for the quantitative synthesis of homogeneity is not surprising, since each subset inall available findings, such analyses in clinical trials offer cludes only trials in which the effects of treatment were an attractive approach to strengthen the evidence about in the same direction. the efficacy of treatment and are, therefore, popular in Table 2 presents treatment effect estimates and the the medical literature. The approach is especially ad- corresponding standard errors for each set of trials vantageous when the individual trials are either too where homogeneity holds. In either scale of measmall or yield inconsistent conclusions about the effect surement, the data from Eriksson et al. indicate that there is no statistically significant effect of treatment on of treatment. 1083
1084
HEPATOLOGY Elsewhere
HEPATOLOGY
TABLE 1. Mortality proportions Eriksson et al. (5)
Naylor et al. Reference ___
Trial
Subset
Treatment
Control
Treatment
H2 H3 H4 H6
I I1 I1 I
5/17 10125 7/23 5/30
8/17 5/25 6/24 16/29
3/17 10125 7/36 15/40
~~
Rossi-Fanelli (1) Wahren (2) Michel (3, 6) Cerra (4)
~~
TABLE 2. Overall treatment effect estimates"and their standard errorsb Naylor et al. Subset
Difference in proportions Log odds
Control
I
I1
Eriksson et al. (5)
-0.31" (0.10)
0.13 (0.09)
0.05 (0.06)
- 1.35'
0.63 (0.46)
0.31 (0.31)
(0.53)
"Estimators from DerSimonian and Laird (7). *Figures in parentheses are the standard errors of the corresponding estimates. 'p < 0.01.
mortality. The same conclusion holds with the data from subset 11. We note that these effect estimates reflect greater (not statistically significant) mortality rates for the treatment group. On the other hand, the data from subset I imply a beneficial effect of treatment on mortality: the estimated difference in mortality proportions between treatment and control groups is -0.31 (with a standard error of O.lO), whereas the log odds ratio estimate is - 1.35 (with a standard error of 0.53). A similar analysis of the data for recovery rates from Eriksson et al. yields the same basic conclusionsas those from Naylor et al.; that is, the homogeneity of the effects of treatment and the beneficial effects of therapy on recovery from hepatic encephalopathy are similar for both studies. These results emphasize the obvious difficulties that arise when collecting data for meta-analysis. Even when meta-analysts consider the same trials with complicated designs, they may abstract drastically different data for the same endpoint and infer conflicting conclusions about treatment efficacy. For meta-analyses of clinical trials, it is clearly preferable to use information available from the most recent and final reports rather than from interim presentations. It is also preferable to include, whenever possible, data on all patients who enter the study. When inconsistencies exist within an individual study because of complicated trial design or inadequate reporting, the robustness of the conclusions from a primary meta-analysis need to be rigorously evaluated through additional analyses with different combinations of the data. Naylor et al. consider several such sensitivity analyses to assess the impact of discrepant data on their conclusions based on primary analyses. Although their primary conclusion of positive treatment effect on mortality holds in all the secondary analyses they
4/17 5/25 7/34 10/35
include, it is not robust with respect to the additional analysis we consider here. In the context of conflicting inferences resulting from discrepancies in data abstraction, trials of sound design, improvement in standards of publication and agreement about the criteria for data abstraction are needed to facilitate consistent data collection. Only then can we strengthen our ability to combine results from clinical trials.
REBECCADERSIMONIAN, S.M., S.D. Department of Public Health Yale University School of Medicine New Haven, Connecticut 06510 REFERENCES 1. Rossi-Fanelli F, Riggio 0, Cangiano C. Branched chain amino acids versus lactulose in the treatment of hepatic coma: a controlled study. Dig Dis Sci 1982;27:929-935. 2. Wahren J , Denis J , Dursurmont P, Eriksson LS, Escoffier J-M, Gauthier AP, Hagenfeldt L, et al. Is intravenous administration of branched chain amino acids effective in the treatment of hep1983;3: atic encephalopathy? A multicenter study. HEPATOLOGY 475-480. 3. Michel H, Pomier-Layrargues G, Aubin JP, Bories P, Mirouze D, Bellet-Hermann H. Treatment of hepatic encephalopathy by infusion of a modified amino-acid solution: results of a controlled study in 47 cirrhotic patients. In: Capocaccia L, Fischer JE, Rossi-Fanelli F, eds. Hepatic encephalopathy in chronic liver failure. New York Plenum, 1984:323-333. 4. Cerra FB, Cheung NK, Fischer JE, Kaplowitz N, Schiff ER, Dienstag JL, Bower RH, et al. Disease-specificamino acid infusion (FO8O) in hepatic encephalopathy: a prospective, randomized, double-blind, controlled trial. J Parenter Enteral Nutr 1985;9:288295. 5. Eriksson LS, Conn HO. Branched-chain amino acids in the management of hepatic encephalopathy: an analysis of variants. HEPATOLOGY 1989;10:228-246. 6. Michel H, Bories P, Aubin JP, Pomier-Layrargues G, Bauret P, Bellet-Herman H. Treatment of acute hepatic encephalopathy in cirrhotics with a branched-chain amino acids enriched versus a conventional amino acids mixture. Liver 1985;5:282-289. 7. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials 1986;7:177-188.
RADIOLOGIC ABLATION OF THE GALLBLADDER: AN ALTERNATIVE TO CHOLECYSTECTOMY IN THE TWENTY-FIRST CENTURY
Becker CD, Quenville NF, Burhenne HJ. Gallbladder ablation through radiologic intervention: an experimental alternative to cholecystectomy. Radiology 1989; 171~235-240.
PARENTERAL NUTRITION WITH BRANCHED-CHAINAMINO ACIDS IN HEPATIC ENCEPHALOPATHY:META ANALYSIS
The report by Naylor et al. not only exemplifies the popularity and attractiveness of the approach, but also underscores some of the problematic issues inherent in Naylor CD, O’Rourke K, Detsky AS, Baker JP. any meta-analysis of clinical trials. This commentary Parenteral nutrition with branched-chain amino acids focuses on one such issue relating to the problem of in hepatic encephalopathy: a meta-analysis. Gastroen- appropriate data abstraction. To evaluate the effect of treatment on mortality, terology 1989;97:1033-1042. Naylor et al. consider four trials (1-4) for “primary” analysis. Table 1 lists the relevant data for these trials EDITOR’S ABSTRACT from the report. This table also lists the corresponding The authors undertook a meta analysis of nine data of a clinical critique of these same trials published randomized clinical trials (RCTs) in which modified by Eriksson and Conn at about the same time (5). Data amino acid solutions that contained supplemental for only one trial (H3) are the same in the two reports. amounts of branched-chain amino acids (BCAA) were Data for the other three trials do not match because of compared with other forms of therapy of hepatic discrepancies in the number of patients studied, the encephalopathy (HE). When five RCTs were pooled, number who survive, or both. The reasons for the analysis showed highly significant improvement in the differences vary: for H2, the “primary” analysis of degree of HE (p < 0.001). Various interpretations of Naylor et al. includes mortality information after the the data gave similar results. The mortality data 48-hour crossover point, whereas Eriksson et al. condiffered so widely in these RCTs that pooling of the sider only the pre-crossover results. For H4, Naylor et al. data could not be meaningfully performed. In two RCTs BCAA therapy resulted in increased mortality, use the data from a preliminary symposium report, whereas in two others BCAA treatment was asso- whereas Eriksson et al. use the data from a subsequent ciated with increased survival. An aggregate relative report ( 6 ) that includes additional patients. For H6, risk reduction of 0.59 was calculated with 95% con- Naylor et al. exclude patients who drop out after fidence intervals of 0.23 to 0.80 (p < 0.002). When the randomization from their “primary” analysis, whereas data from one additional study that had not yet been Eriksson et al. consider all patients who enter the study. published and that showed BCAA therapy to be beneUsing the methods of DerSimonian and Laird (7) to ficial was added to the analysis, the pooled relative evaluate homogeneity of treatment effect, we find that reduction was even greater (i.e., 0.82 [p < 0.00011). The the data from Naylor et al. indicate a large degree mortality in these studies was significantly reduced of heterogeneity (p < 0.021, whereas the data from regardless of the type of statistical analysis performed (p = 0.023). Nevertheless, because of the short du- Eriksson et al. strongly imply homogeneity. These ration of follow-up in these studies and the striking findings hold both when treatment effect is the difdifferences in the effects of BCAA on mortality, the ference in proportions and when it is the log odds ratio. In light of the heterogeneity in the results of authors recommended that additional RCTs with protreatment in the four studies, Naylor et al. sought longed follow-up be performed. explanations for the disparities. Since the trials varied widely in the duration of therapy and in the intake of COMMENTS calories, they divided the trials into two subsets of Meta-analysis is a general term to describe the similar length of follow-up and caloric intake with two statistical analysis of a group of results of individual trials each (Table 1).Within each subset, the homogeinvestigations for the purpose of integrating the find- neity of results of therapy was more homogeneous. This ings. Since they allow for the quantitative synthesis of homogeneity is not surprising, since each subset inall available findings, such analyses in clinical trials offer cludes only trials in which the effects of treatment were an attractive approach to strengthen the evidence about in the same direction. the efficacy of treatment and are, therefore, popular in Table 2 presents treatment effect estimates and the the medical literature. The approach is especially ad- corresponding standard errors for each set of trials vantageous when the individual trials are either too where homogeneity holds. In either scale of measmall or yield inconsistent conclusions about the effect surement, the data from Eriksson et al. indicate that there is no statistically significant effect of treatment on of treatment. 1083
1084
HEPATOLOGY Elsewhere
HEPATOLOGY
TABLE 1. Mortality proportions Eriksson et al. (5)
Naylor et al. Reference ___
Trial
Subset
Treatment
Control
Treatment
H2 H3 H4 H6
I I1 I1 I
5/17 10125 7/23 5/30
8/17 5/25 6/24 16/29
3/17 10125 7/36 15/40
~~
Rossi-Fanelli (1) Wahren (2) Michel (3, 6) Cerra (4)
~~
TABLE 2. Overall treatment effect estimates"and their standard errorsb Naylor et al. Subset
Difference in proportions Log odds
Control
I
I1
Eriksson et al. (5)
-0.31" (0.10)
0.13 (0.09)
0.05 (0.06)
- 1.35'
0.63 (0.46)
0.31 (0.31)
(0.53)
"Estimators from DerSimonian and Laird (7). *Figures in parentheses are the standard errors of the corresponding estimates. 'p < 0.01.
mortality. The same conclusion holds with the data from subset 11. We note that these effect estimates reflect greater (not statistically significant) mortality rates for the treatment group. On the other hand, the data from subset I imply a beneficial effect of treatment on mortality: the estimated difference in mortality proportions between treatment and control groups is -0.31 (with a standard error of O.lO), whereas the log odds ratio estimate is - 1.35 (with a standard error of 0.53). A similar analysis of the data for recovery rates from Eriksson et al. yields the same basic conclusionsas those from Naylor et al.; that is, the homogeneity of the effects of treatment and the beneficial effects of therapy on recovery from hepatic encephalopathy are similar for both studies. These results emphasize the obvious difficulties that arise when collecting data for meta-analysis. Even when meta-analysts consider the same trials with complicated designs, they may abstract drastically different data for the same endpoint and infer conflicting conclusions about treatment efficacy. For meta-analyses of clinical trials, it is clearly preferable to use information available from the most recent and final reports rather than from interim presentations. It is also preferable to include, whenever possible, data on all patients who enter the study. When inconsistencies exist within an individual study because of complicated trial design or inadequate reporting, the robustness of the conclusions from a primary meta-analysis need to be rigorously evaluated through additional analyses with different combinations of the data. Naylor et al. consider several such sensitivity analyses to assess the impact of discrepant data on their conclusions based on primary analyses. Although their primary conclusion of positive treatment effect on mortality holds in all the secondary analyses they
4/17 5/25 7/34 10/35
include, it is not robust with respect to the additional analysis we consider here. In the context of conflicting inferences resulting from discrepancies in data abstraction, trials of sound design, improvement in standards of publication and agreement about the criteria for data abstraction are needed to facilitate consistent data collection. Only then can we strengthen our ability to combine results from clinical trials.
REBECCADERSIMONIAN, S.M., S.D. Department of Public Health Yale University School of Medicine New Haven, Connecticut 06510 REFERENCES 1. Rossi-Fanelli F, Riggio 0, Cangiano C. Branched chain amino acids versus lactulose in the treatment of hepatic coma: a controlled study. Dig Dis Sci 1982;27:929-935. 2. Wahren J , Denis J , Dursurmont P, Eriksson LS, Escoffier J-M, Gauthier AP, Hagenfeldt L, et al. Is intravenous administration of branched chain amino acids effective in the treatment of hep1983;3: atic encephalopathy? A multicenter study. HEPATOLOGY 475-480. 3. Michel H, Pomier-Layrargues G, Aubin JP, Bories P, Mirouze D, Bellet-Hermann H. Treatment of hepatic encephalopathy by infusion of a modified amino-acid solution: results of a controlled study in 47 cirrhotic patients. In: Capocaccia L, Fischer JE, Rossi-Fanelli F, eds. Hepatic encephalopathy in chronic liver failure. New York Plenum, 1984:323-333. 4. Cerra FB, Cheung NK, Fischer JE, Kaplowitz N, Schiff ER, Dienstag JL, Bower RH, et al. Disease-specificamino acid infusion (FO8O) in hepatic encephalopathy: a prospective, randomized, double-blind, controlled trial. J Parenter Enteral Nutr 1985;9:288295. 5. Eriksson LS, Conn HO. Branched-chain amino acids in the management of hepatic encephalopathy: an analysis of variants. HEPATOLOGY 1989;10:228-246. 6. Michel H, Bories P, Aubin JP, Pomier-Layrargues G, Bauret P, Bellet-Herman H. Treatment of acute hepatic encephalopathy in cirrhotics with a branched-chain amino acids enriched versus a conventional amino acids mixture. Liver 1985;5:282-289. 7. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials 1986;7:177-188.
RADIOLOGIC ABLATION OF THE GALLBLADDER: AN ALTERNATIVE TO CHOLECYSTECTOMY IN THE TWENTY-FIRST CENTURY
Becker CD, Quenville NF, Burhenne HJ. Gallbladder ablation through radiologic intervention: an experimental alternative to cholecystectomy. Radiology 1989; 171~235-240.
