- Email: [email protected]
Dietary sodium intake and mortality: NHANES
CORRESPONDENCE
Frequency of cystic fibrosis ⌬F508 heterozygotes (%)
9 p=0·02
FEV1 % predicted
p=0·005
6 p=0·03 NS
3
NS
NS
0 9 p=0·003 FVC % predicted
6
p=0·02
NS
NS
NS
3
0
NS
⬍1 ⬍2 ⬍5 ⬍10 ⬍30 ⬍50 ⭓50
Percentiles in total general population sample Frequency of ⌬F508 heterozygotes within corresponding percentile group versus frequency of ⌬F508 heterozygotes within remaining sample* *Lowest 2% vs highest 98%; 2-test, NS=p>0·05.
pulmonary disease (COPD) are similar to those recommended by others,1,2 but not mutually exclusive. Asthma and chronic bronchitis rely on affirmative answers to questionnaires (table 1 in our report), whereas COPD is diagnosed based on spirometry as airway obstruction excluding asthmatics. In epidemiological surveys, self-reported asthma seems the best way to assess asthma.2 Number of individuals with asthma, chronic bronchitis, and/or COPD did not differ between ⌬F508 carriers and non-carriers (57, [23%] of 250 vs 2070 [23%] of 8841; 2 test p=0·82), but this does not invalidate the finding that ⌬F508 carriers reported asthma more often than non-carriers. The 17 180 individuals were drawn randomly from the Copenhagen Central Population Register to obtain a representative sample of the general population. Since only 53% of invited individuals participated in our study, selection bias may have been present, as in any other study. By contrast with the suggestion by Michael Swift and Yun Su, chronic airway disease is more common among non-responders than among responders.3 Therefore, and since ⌬F508 carriers were clustered among those with the poorest lung function (figure), this selection bias probably caused us to underestimate the effect of ⌬F508. Since the incidence of cystic fibrosis in Denmark is 1 in 4700, knowledge of cystic
THE LANCET • Vol 352 • September 19, 1998
fibrosis in families is unlikely to affect our results. We state that “individuals heterozygous with ⌬F508 may account for up to 2% of people with asthma”. This estimate of attributable risk approximates to the excess fraction of asthma-cases that would not have occurred had the ⌬F508 exposure been absent. Among individuals with asthma, 4·2% (not 2%) were ⌬F508 heterozygotes compared with 2·7% among individuals without asthma. John Warner, Albert Lowenfels, F Kaufmann, and their colleagues provide evidence for an increased prevalence of allergy among cycstic fibrosis heterozygotes. Participants in our study were asked: “does exposure to food, medicine, flowers, animals or anything else cause asthma”. Of ⌬F508 carriers, 17 (7%) of 250 answered “yes” compared with 389 (4%) of 8780 non-carriers (2 test p=0·08). Lowenfels and co-workers discuss selective advantage among cystic fibrosis heterozygotes, which may explain the remarkably high incidence of cystic fibrosis among white people. We showed that among non-smokers ⌬F508 heterozygotes have a reproductive advantage.4 On the basis of a study of 73 ⌬F508 carriers and 304 relatives, Swift and Su conclude that ⌬F508 carriers may be protected from asthma. We provide evidence of the opposite in our report. Our study has the advantage that we collected and genotyped a large general population sample (n=9141) without previous knowledge of cystic fibrosis in the families, thus avoiding ascertainment bias. Morten Dahl, Anne Tybjaerg-Hansen, Peter Lange, *Børge G Nordestgaard Department of Clinical Biochemistry, Glostrup University Hospital, DK-2600 Glostrup, Denmark (e-mail: [email protected]) 1
2
3
4
British Society guidelines for the management of chronic obstructive pulmonary disease. Thorax 1997; 52: (suppl 5). Enarson DA, Vedal S, Schulzer M, Dybuncio A, Chan-Yeung M. Asthma, asthma-like symptoms, chronic bronchitis, and the degree of bronchial hyperresponsiveness in epidemiologic surveys. Am Rev Respir Dis 1987; 136: 613–17. Lange P. Development and prognosis of chronic obstructive pulmonary disease with special reference to the role of tobacco smoking. Dan Med Bull 1992; 39; 30–48. Dahl M, Tybjaerg-Hansen A, Wittrup HH, Lange P, Nordestgaard BG. Cystic fibrosis ⌬F508 heterozygotes, smoking, and reproduction: studies of 9141 individuals from a general population sample. Genomics 1998; 50: 89–96.
Dietary sodium intake and mortality: NHANES Sir—The analyses of data from the National Health and Nutrition Examination Survey (NHANES I), reported by Michael Alderman and colleagues (March 14, p 781)1 are seriously flawed. The analyses as presented do not exclude individuals with known cardiovascular disease (CVD) or hypertension, who were disproportionately represented in the lowest sodium quartile. Such individuals usually report lower sodium intake and have increased risk of fatal coronary heart disease. Analyses excluding participants with a history of CVD (but apparently not those with a history of hypertension) were reportedly similar but should have been presented. Participants reportedly in the lowest quartile of sodium intake not only had the highest prevalence of known CVD and history of hypertension but were also the oldest, had the highest proportion of black participants, and had the lowest body weights, all of which, except sodium intake, are associated with increased death rates. It would therefore be expected that, irrespective of sodium intake, those in the lowest quartile would exhibit the highest mortality. The inverse association between blood pressure and reported sodium intake conflicts with most previous publications.2 However, reported calorie intake was also inversely associated with blood pressure and mortality. Most nutrients in the diet increase with increasing calorie intake. Thus, to avoid confounding by low calorie intake, the pertinent analyses were those in which sodium intake was corrected for dietary calories. Data on smoking were collected, but not included in analyses. Because smokers have different diets,3 lower body weights and blood pressures,4 are from lower socioeconomic strata, and have higher death rates than nonsmokers, this omission makes serious confounding likely. The inclusion in the multivariate analyses of variables such as sodium intake and blood pressure, which may well be related in the causal pathway result in over-fitting the model, can critically change the size and even the direction of the effect under investigation. These errors notwithstanding, the investigators misinterpret their findings. For example, the reported relative risk for all-cause mortality (RR) associated with a 1313 mg difference in sodium intake was 0·88. However, using their data to estimate the effect of a 1313 mg
987
CORRESPONDENCE
difference in sodium intake in their model, one must assume some specific individual values of calorie and of sodium intake. If we compute the RR of two people with sodium intakes of 4000 mg and 2700 mg, respectively, and the same calorie intake of 2000 kcal (8360 kJ), the estimated RR is actually 1·00. A similar computation with CVD mortality as the outcome, gives an estimated RR of 1·02 rather than 0·89, as Alderman and colleagues reported. Despite all these major methodological flaws, higher sodium per calories intake was significantly associated with higher rates of cardiovascular (p=0·0056) and all-cause mortality (p=0·0004). It is hard to understand why the investigators discount these results as unsupportive of recommendations to use foods with lower salt content. If following appropriate analyses of their NHANES data, the findings were to remain, it would then be important to attempt to replicate the findings in other studies, given that most previously published data have shown a direct association between sodium intake and blood pressure2 and that all low-bloodpressure populations have vanishingly low rates of CVD.5 N R Poulter on behalf of The Faculty 31st International Society and Federation of Cardiology 10-day Teaching Seminar in Cardiovascular Disease, Epidemiology and Prevention Cardiovascular Studies Unit, Department of Clinical Pharmacology, Imperial College School of Medicine, London W2 1PG, UK The faculty consisted of: Elizabeth Barret-Connor, Edward C Davis, Vinjar Fønnebø, Kay-Tee Khaw, Darwin R Labarthe, Neil R Poulter, K Srinath Reddy, Grethe S Tell, Dag S Thelle 1
2
3
4
5
Alderman MH, Cohen H, Madhavan S. Dietary sodium intake and mortality: the National Health and Nutrition Examination Survey (NHANES I). Lancet 1998; 1: 781–85. Frost CD, Law MR, Wald NJ. By how much does dietary salt reduction lower blood pressure? II: analysis of observational data within populations. BMJ 1991; 302: 815–18. Margetts BM, Jackson AA. Interactions between people’s diet and their smoking habits: the dietary and nutritional survey of British adults. BMJ 1993; 307: 1381–84. Wolf A, D’Agostino RB, Kannel WB, Bonita R, Belanger AJ. Cigarette smoking as a risk factor for stroke: the Framingham Study. JAMA 1988; 259: 1025–29. Lowenstein FW. Blood pressure in relation to age and sex in the tropics and subtropics: a review of the literature and an investigation in two tribes of Brazil Indians. Lancet 1961; 1: 389–92.
Authors’ reply Sir—Neil Poulter and colleagues suggest that further analyses might alter the
988
inverse association of sodium to mortality found in the NHANES I study. Application of their suggestions do not alter the published findings. As suggested, we excluded individuals with known CVD. The age-adjusted and sex-adjusted rates per 1000 personyears were 21·26 and 17·53 for the lowest and highest sodium quartiles (p=0·0001) for all-cause, and 10·55 and 8·30 (p=0·0011) for CVD mortality. Corresponding values, excluding a history of CVD and hypertension, were 19·92 and 16·25 (p=0·0003) for all cause, and 9·65 and 7·59 (p-0·0063) for CVD mortality. The hazard ratios of all-cause mortality and CVD mortality for 1 SD sodium found when those with history of hypertension or CVD were excluded (n-8508) are 0·91 (95% CI 0·81–1·03) and 0·90 (0·75–1·08) with the full Cox’s model. Corresponding values with the best-fit model using stepwise elimination with greater than 0·100 criterion are 0·88 (0·82–0·93) and 0·88 (0·80–0·97). Poulter noted the seemingly inverse association of sodium intake and blood pressure. This finding actually reflects confounding. When age, ethnic origin, and body-mass index, along with history of CVD, and hypertension, were controlled for in a linear regression model, there were no significant associations of blood pressure with sodium, calories, or their ratio (p=0·587, 0·502, 0·499, respectively). Smoking could not be assessed, since baseline data were available only for 20% of patients. In a previous report, smoking, significantly associated with outcome, did not confound the low sodium to CVD association.1 Repeating the published Cox models without systolic blood pressure did not alter the hazard ratios for sodium. Poulter provides a specific example of a 1300 mg sodium difference with constant calories to test the combined effect of sodium and sodium/calorie ratio. In his example, there is no evidence of benefit or harm—the relative risk is 1. However, constant calories is an extreme assumption. The alternate extreme assumption is a constant sodium/calorie ratio (in his example, 2), by which the 2700 mg sodium in the example would conform to a calorie intake of 5643 kJ. Using Poulter’s computing method, the relative risk is now 0·88. In practice, of course the calorie variation would likely fall between these two extremes, so that the combined risk of 4000 mg versus 2700 mg sodium would be between 0·88 and 1—in other words never associated with a benefit for lower sodium. Finally, Poulter reminds us of the
association of sodium with blood pressure in Brazilian Indians, and implies that this produces lower CVD. If ecological argument is useful, one might note the Japanese, with a high salt intake, enjoy a life expectancy among the world’s longest. Variation in sodium intake has many effects. The combination of all these effecs will determine the effects of sodium on health. The further analyses suggested by Poulter reaffirm our original findings. We heartily endorse his call for further data to confirm whether the inverse association of sodium to mortality seen here is found in other populations. *Michael H Alderman, Hillel W Cohen, Shantha Madhavan Division of Epidemiology and Social Medicine, Albert Einstein College of Medicine, Bronx, NY 10461, USA 1
Alderman MH, Madhavan S, Cohen H, Sealey JE, Laragh JH. Low urinary sodium is associated with greater risk of myocardial infarction among treated hypertensive men. Hypertension 1995; 25: 1144–52.
Prognostic value of Th1/Th2 ratio in rheumatoid arthritis Sir—W L von der Graaff and colleagues (June 27, p 1931)1 report a changing ratio of Th1/Th2 cell subsets in the peripheral blood of patients with rheumatoid arthritis. Although their report is stimulating, it raises several questions. First, it may be inconclusive to determine the Th1/Th2 subsets in human beings from only intracellular staining of interferon-␥ (IFN␥) and interleukin (IL)4, without investigation of other cytokines such as IL2, IL10, or IL5. For example, as previously reported, Th0 cells might be included in peripheral blood of patients with rheumatoid arthritis. Second, since the data only showed the ratio of, but not number of, Th1/Th2 cells, it is unclear whether the decreased Th1 or increased Th2 cells would be primarily involved in the reported better outcome of rheumatoid arthritis after 9 months of therapy. Third, the investigators found “large differences in Th1/Th2 ratios between patients” with no correlation with disease activity before treatment, and also a change of Th1/Th2 ratio during the course of therapy. The figure, however, showed the ratio only before treatment and disease activity score after 9 months of therapy. Thus, the disease activity scores shown in the figure do not directly verify whether the treatment was effective enough to decrease the score in each patient. This question may also
THE LANCET • Vol 352 • September 19, 1998
Frequency of cystic fibrosis ⌬F508 heterozygotes (%)
9 p=0·02
FEV1 % predicted
p=0·005
6 p=0·03 NS
3
NS
NS
0 9 p=0·003 FVC % predicted
6
p=0·02
NS
NS
NS
3
0
NS
⬍1 ⬍2 ⬍5 ⬍10 ⬍30 ⬍50 ⭓50
Percentiles in total general population sample Frequency of ⌬F508 heterozygotes within corresponding percentile group versus frequency of ⌬F508 heterozygotes within remaining sample* *Lowest 2% vs highest 98%; 2-test, NS=p>0·05.
pulmonary disease (COPD) are similar to those recommended by others,1,2 but not mutually exclusive. Asthma and chronic bronchitis rely on affirmative answers to questionnaires (table 1 in our report), whereas COPD is diagnosed based on spirometry as airway obstruction excluding asthmatics. In epidemiological surveys, self-reported asthma seems the best way to assess asthma.2 Number of individuals with asthma, chronic bronchitis, and/or COPD did not differ between ⌬F508 carriers and non-carriers (57, [23%] of 250 vs 2070 [23%] of 8841; 2 test p=0·82), but this does not invalidate the finding that ⌬F508 carriers reported asthma more often than non-carriers. The 17 180 individuals were drawn randomly from the Copenhagen Central Population Register to obtain a representative sample of the general population. Since only 53% of invited individuals participated in our study, selection bias may have been present, as in any other study. By contrast with the suggestion by Michael Swift and Yun Su, chronic airway disease is more common among non-responders than among responders.3 Therefore, and since ⌬F508 carriers were clustered among those with the poorest lung function (figure), this selection bias probably caused us to underestimate the effect of ⌬F508. Since the incidence of cystic fibrosis in Denmark is 1 in 4700, knowledge of cystic
THE LANCET • Vol 352 • September 19, 1998
fibrosis in families is unlikely to affect our results. We state that “individuals heterozygous with ⌬F508 may account for up to 2% of people with asthma”. This estimate of attributable risk approximates to the excess fraction of asthma-cases that would not have occurred had the ⌬F508 exposure been absent. Among individuals with asthma, 4·2% (not 2%) were ⌬F508 heterozygotes compared with 2·7% among individuals without asthma. John Warner, Albert Lowenfels, F Kaufmann, and their colleagues provide evidence for an increased prevalence of allergy among cycstic fibrosis heterozygotes. Participants in our study were asked: “does exposure to food, medicine, flowers, animals or anything else cause asthma”. Of ⌬F508 carriers, 17 (7%) of 250 answered “yes” compared with 389 (4%) of 8780 non-carriers (2 test p=0·08). Lowenfels and co-workers discuss selective advantage among cystic fibrosis heterozygotes, which may explain the remarkably high incidence of cystic fibrosis among white people. We showed that among non-smokers ⌬F508 heterozygotes have a reproductive advantage.4 On the basis of a study of 73 ⌬F508 carriers and 304 relatives, Swift and Su conclude that ⌬F508 carriers may be protected from asthma. We provide evidence of the opposite in our report. Our study has the advantage that we collected and genotyped a large general population sample (n=9141) without previous knowledge of cystic fibrosis in the families, thus avoiding ascertainment bias. Morten Dahl, Anne Tybjaerg-Hansen, Peter Lange, *Børge G Nordestgaard Department of Clinical Biochemistry, Glostrup University Hospital, DK-2600 Glostrup, Denmark (e-mail: [email protected]) 1
2
3
4
British Society guidelines for the management of chronic obstructive pulmonary disease. Thorax 1997; 52: (suppl 5). Enarson DA, Vedal S, Schulzer M, Dybuncio A, Chan-Yeung M. Asthma, asthma-like symptoms, chronic bronchitis, and the degree of bronchial hyperresponsiveness in epidemiologic surveys. Am Rev Respir Dis 1987; 136: 613–17. Lange P. Development and prognosis of chronic obstructive pulmonary disease with special reference to the role of tobacco smoking. Dan Med Bull 1992; 39; 30–48. Dahl M, Tybjaerg-Hansen A, Wittrup HH, Lange P, Nordestgaard BG. Cystic fibrosis ⌬F508 heterozygotes, smoking, and reproduction: studies of 9141 individuals from a general population sample. Genomics 1998; 50: 89–96.
Dietary sodium intake and mortality: NHANES Sir—The analyses of data from the National Health and Nutrition Examination Survey (NHANES I), reported by Michael Alderman and colleagues (March 14, p 781)1 are seriously flawed. The analyses as presented do not exclude individuals with known cardiovascular disease (CVD) or hypertension, who were disproportionately represented in the lowest sodium quartile. Such individuals usually report lower sodium intake and have increased risk of fatal coronary heart disease. Analyses excluding participants with a history of CVD (but apparently not those with a history of hypertension) were reportedly similar but should have been presented. Participants reportedly in the lowest quartile of sodium intake not only had the highest prevalence of known CVD and history of hypertension but were also the oldest, had the highest proportion of black participants, and had the lowest body weights, all of which, except sodium intake, are associated with increased death rates. It would therefore be expected that, irrespective of sodium intake, those in the lowest quartile would exhibit the highest mortality. The inverse association between blood pressure and reported sodium intake conflicts with most previous publications.2 However, reported calorie intake was also inversely associated with blood pressure and mortality. Most nutrients in the diet increase with increasing calorie intake. Thus, to avoid confounding by low calorie intake, the pertinent analyses were those in which sodium intake was corrected for dietary calories. Data on smoking were collected, but not included in analyses. Because smokers have different diets,3 lower body weights and blood pressures,4 are from lower socioeconomic strata, and have higher death rates than nonsmokers, this omission makes serious confounding likely. The inclusion in the multivariate analyses of variables such as sodium intake and blood pressure, which may well be related in the causal pathway result in over-fitting the model, can critically change the size and even the direction of the effect under investigation. These errors notwithstanding, the investigators misinterpret their findings. For example, the reported relative risk for all-cause mortality (RR) associated with a 1313 mg difference in sodium intake was 0·88. However, using their data to estimate the effect of a 1313 mg
987
CORRESPONDENCE
difference in sodium intake in their model, one must assume some specific individual values of calorie and of sodium intake. If we compute the RR of two people with sodium intakes of 4000 mg and 2700 mg, respectively, and the same calorie intake of 2000 kcal (8360 kJ), the estimated RR is actually 1·00. A similar computation with CVD mortality as the outcome, gives an estimated RR of 1·02 rather than 0·89, as Alderman and colleagues reported. Despite all these major methodological flaws, higher sodium per calories intake was significantly associated with higher rates of cardiovascular (p=0·0056) and all-cause mortality (p=0·0004). It is hard to understand why the investigators discount these results as unsupportive of recommendations to use foods with lower salt content. If following appropriate analyses of their NHANES data, the findings were to remain, it would then be important to attempt to replicate the findings in other studies, given that most previously published data have shown a direct association between sodium intake and blood pressure2 and that all low-bloodpressure populations have vanishingly low rates of CVD.5 N R Poulter on behalf of The Faculty 31st International Society and Federation of Cardiology 10-day Teaching Seminar in Cardiovascular Disease, Epidemiology and Prevention Cardiovascular Studies Unit, Department of Clinical Pharmacology, Imperial College School of Medicine, London W2 1PG, UK The faculty consisted of: Elizabeth Barret-Connor, Edward C Davis, Vinjar Fønnebø, Kay-Tee Khaw, Darwin R Labarthe, Neil R Poulter, K Srinath Reddy, Grethe S Tell, Dag S Thelle 1
2
3
4
5
Alderman MH, Cohen H, Madhavan S. Dietary sodium intake and mortality: the National Health and Nutrition Examination Survey (NHANES I). Lancet 1998; 1: 781–85. Frost CD, Law MR, Wald NJ. By how much does dietary salt reduction lower blood pressure? II: analysis of observational data within populations. BMJ 1991; 302: 815–18. Margetts BM, Jackson AA. Interactions between people’s diet and their smoking habits: the dietary and nutritional survey of British adults. BMJ 1993; 307: 1381–84. Wolf A, D’Agostino RB, Kannel WB, Bonita R, Belanger AJ. Cigarette smoking as a risk factor for stroke: the Framingham Study. JAMA 1988; 259: 1025–29. Lowenstein FW. Blood pressure in relation to age and sex in the tropics and subtropics: a review of the literature and an investigation in two tribes of Brazil Indians. Lancet 1961; 1: 389–92.
Authors’ reply Sir—Neil Poulter and colleagues suggest that further analyses might alter the
988
inverse association of sodium to mortality found in the NHANES I study. Application of their suggestions do not alter the published findings. As suggested, we excluded individuals with known CVD. The age-adjusted and sex-adjusted rates per 1000 personyears were 21·26 and 17·53 for the lowest and highest sodium quartiles (p=0·0001) for all-cause, and 10·55 and 8·30 (p=0·0011) for CVD mortality. Corresponding values, excluding a history of CVD and hypertension, were 19·92 and 16·25 (p=0·0003) for all cause, and 9·65 and 7·59 (p-0·0063) for CVD mortality. The hazard ratios of all-cause mortality and CVD mortality for 1 SD sodium found when those with history of hypertension or CVD were excluded (n-8508) are 0·91 (95% CI 0·81–1·03) and 0·90 (0·75–1·08) with the full Cox’s model. Corresponding values with the best-fit model using stepwise elimination with greater than 0·100 criterion are 0·88 (0·82–0·93) and 0·88 (0·80–0·97). Poulter noted the seemingly inverse association of sodium intake and blood pressure. This finding actually reflects confounding. When age, ethnic origin, and body-mass index, along with history of CVD, and hypertension, were controlled for in a linear regression model, there were no significant associations of blood pressure with sodium, calories, or their ratio (p=0·587, 0·502, 0·499, respectively). Smoking could not be assessed, since baseline data were available only for 20% of patients. In a previous report, smoking, significantly associated with outcome, did not confound the low sodium to CVD association.1 Repeating the published Cox models without systolic blood pressure did not alter the hazard ratios for sodium. Poulter provides a specific example of a 1300 mg sodium difference with constant calories to test the combined effect of sodium and sodium/calorie ratio. In his example, there is no evidence of benefit or harm—the relative risk is 1. However, constant calories is an extreme assumption. The alternate extreme assumption is a constant sodium/calorie ratio (in his example, 2), by which the 2700 mg sodium in the example would conform to a calorie intake of 5643 kJ. Using Poulter’s computing method, the relative risk is now 0·88. In practice, of course the calorie variation would likely fall between these two extremes, so that the combined risk of 4000 mg versus 2700 mg sodium would be between 0·88 and 1—in other words never associated with a benefit for lower sodium. Finally, Poulter reminds us of the
association of sodium with blood pressure in Brazilian Indians, and implies that this produces lower CVD. If ecological argument is useful, one might note the Japanese, with a high salt intake, enjoy a life expectancy among the world’s longest. Variation in sodium intake has many effects. The combination of all these effecs will determine the effects of sodium on health. The further analyses suggested by Poulter reaffirm our original findings. We heartily endorse his call for further data to confirm whether the inverse association of sodium to mortality seen here is found in other populations. *Michael H Alderman, Hillel W Cohen, Shantha Madhavan Division of Epidemiology and Social Medicine, Albert Einstein College of Medicine, Bronx, NY 10461, USA 1
Alderman MH, Madhavan S, Cohen H, Sealey JE, Laragh JH. Low urinary sodium is associated with greater risk of myocardial infarction among treated hypertensive men. Hypertension 1995; 25: 1144–52.
Prognostic value of Th1/Th2 ratio in rheumatoid arthritis Sir—W L von der Graaff and colleagues (June 27, p 1931)1 report a changing ratio of Th1/Th2 cell subsets in the peripheral blood of patients with rheumatoid arthritis. Although their report is stimulating, it raises several questions. First, it may be inconclusive to determine the Th1/Th2 subsets in human beings from only intracellular staining of interferon-␥ (IFN␥) and interleukin (IL)4, without investigation of other cytokines such as IL2, IL10, or IL5. For example, as previously reported, Th0 cells might be included in peripheral blood of patients with rheumatoid arthritis. Second, since the data only showed the ratio of, but not number of, Th1/Th2 cells, it is unclear whether the decreased Th1 or increased Th2 cells would be primarily involved in the reported better outcome of rheumatoid arthritis after 9 months of therapy. Third, the investigators found “large differences in Th1/Th2 ratios between patients” with no correlation with disease activity before treatment, and also a change of Th1/Th2 ratio during the course of therapy. The figure, however, showed the ratio only before treatment and disease activity score after 9 months of therapy. Thus, the disease activity scores shown in the figure do not directly verify whether the treatment was effective enough to decrease the score in each patient. This question may also
THE LANCET • Vol 352 • September 19, 1998