Treating individuals 2. Subgroup analysis in randomised controlled trials: importance, indications, and interpretation

Testing multiple statistical hypotheses resulted in spurious associations: a study of astrological signs and health

OBJECTIVES

To illustrate how multiple hypotheses testing can produce associations with no clinical plausibility.

STUDY DESIGN AND SETTING

We conducted a study of all 10,674,945 residents of Ontario aged between 18 and 100 years in 2000. Residents were randomly assigned to equally sized derivation and validation cohorts and classified according to their astrological sign. Using the derivation cohort, we searched through 223 of the most common diagnoses for hospitalization until we identified two for which subjects born under one astrological sign had a significantly higher probability of hospitalization compared to subjects born under the remaining signs combined (P<0.05).

RESULTS

We tested these 24 associations in the independent validation cohort. Residents born under Leo had a higher probability of gastrointestinal hemorrhage (P=0.0447), while Sagittarians had a higher probability of humerus fracture (P=0.0123) compared to all other signs combined. After adjusting the significance level to account for multiple comparisons, none of the identified associations remained significant in either the derivation or validation cohort.

CONCLUSIONS

Our analyses illustrate how the testing of multiple, non-prespecified hypotheses increases the likelihood of detecting implausible associations. Our findings have important implications for the analysis and interpretation of clinical studies.

Effect of Different Partogram Action Lines on Birth Outcomes

OBJECTIVE

The World Health Organization recommends partograms with a 4-hour action line, denoting the timing of intervention for prolonged labor; others recommend earlier intervention. We assessed the effect of different action line positioning on birth outcomes.

METHODS

A randomized trial of primigravid women with uncomplicated pregnancies, in spontaneous labor at term, was conducted in the northwest of England. Women were assigned to have their labors recorded on a partogram with an action line 2 or 4 hours to the right of the alert line. If progress crossed the action line, diagnosis of prolonged labor was made and managed according to standard protocol. Primary outcomes were rate of cesarean delivery and maternal satisfaction.

RESULTS

A total of 3,000 women were randomly assigned to groups; 2,975 (99.2%) were available for analysis. Questionnaires were completed by 1,929 (65%) women. There were no differences in cesarean delivery rate (136/1,490 compared with 135/1,485; relative risk [RR] 1, 95% confidence interval [CI] 0.80-1.26) or women dissatisfied with labor experience (72/962 compared with 81/967; RR 0.89, 95% CI 0.66-1.21). More women assigned to the 2-hour arm had labors that crossed the action line (854/1,490 compared with 673/1,485; RR 1.27, 95% CI 1.18-1.37); received more intervention (772/1,490 compared with 624/1,485; RR 1.23, 95% CI 1.14-1.33); and, if admitted to the midwife-led unit, were transferred for consultant-led care (366/674 compared with 285/666; RR 1.26, 95% CI 1.13-1.42).

CONCLUSION

In this birth setting, for primigravid women selecting low intervention care, the 2-hour partogram increases the need for intervention without improving maternal or neonatal outcomes, compared with the 4-hour partogram, advocated by the World Health Organization.

CLINICAL TRIAL REGISTRATION

Current Controlled Trials, http://www.controlled-trials.com/isrctn/trial/|/0/78346801.html, ISRCTN78346801.

Continuous cardiotocography (CTG) as a form of electronic fetal monitoring (EFM) for fetal assessment during labour

BACKGROUND

Cardiotocography (sometimes known as electronic fetal monitoring), records changes in the fetal heart rate and their temporal relationship to uterine contractions. The aim is to identify babies who may be short of oxygen (hypoxic), so additional assessments of fetal well-being may be used, or the baby delivered by caesarean section or instrumental vaginal birth.

OBJECTIVES

To evaluate the effectiveness of continuous cardiotocography during labour.

SEARCH STRATEGY

We searched the Cochrane Pregnancy and Childbirth Group Trials Register (March 2006), CENTRAL (The Cochrane Library 2005, Issue 4), MEDLINE (1966 to December 2005), EMBASE (1974 to December 2005), Dissertation Abstracts (1980 to December 2005) and the National Research Register (December 2005).

SELECTION CRITERIA

Randomised and quasi-randomised controlled trials involving a comparison of continuous cardiotocography (with and without fetal blood sampling) with (a) no fetal monitoring, (b) intermittent auscultation (c) intermittent cardiotocography.

DATA COLLECTION AND ANALYSIS

Two authors independently assessed eligibility, quality and extracted data.

MAIN RESULTS

Twelve trials were included (over 37,000 women); only two were high quality. Compared to intermittent auscultation, continuous cardiotocography showed no significant difference in overall perinatal death rate (relative risk (RR) 0.85, 95% confidence interval (CI) 0.59 to 1.23, n = 33,513, 11 trials), but was associated with a halving of neonatal seizures (RR 0.50, 95% CI 0.31 to 0.80, n = 32,386, nine trials) although no significant difference was detected in cerebral palsy (RR 1.74, 95% CI 0.97 to 3.11, n = 13,252, two trials). There was a significant increase in caesarean sections associated with continuous cardiotocography (RR 1.66, 95% CI 1.30 to 2.13, n =18,761, 10 trials). Women were also more likely to have an instrumental vaginal birth (RR 1.16, 95% CI 1.01 to 1.32, n = 18,151, nine trials). Data for subgroups of low-risk, high-risk, preterm pregnancies and high quality trials were consistent with overall results. Access to fetal blood sampling did not appear to influence the difference in neonatal seizures nor any other prespecified outcome.

AUTHORS' CONCLUSIONS

Continuous cardiotocography during labour is associated with a reduction in neonatal seizures, but no significant differences in cerebral palsy, infant mortality or other standard measures of neonatal well-being. However, continuous cardiotocography was associated with an increase in caesarean sections and instrumental vaginal births. The real challenge is how best to convey this uncertainty to women to enable them to make an informed choice without compromising the normality of labour.

Meta-analysis in the assessment of treatment outcome

Randomized clinical trials provide the most reliable estimates of the benefits and harms of treatments. Limited sample sizes restrict their power to allow informative analyses of secondary outcomes, or patient subgroups. The overall results of trials only apply to the average patient and clinical application ignores the individual patient differences.Meta-analysis in the context of a systematic review can produce more precise estimates of effect by combining the results of primary studies. This is particularly valuable for investigating rare, but important outcomes such as suicide. Variations between the trial-specific results can be investigated by meta-regression. Individual patient data meta-analyses (IPDMAs) are potentially much more powerful designs because they allow analysis of patient-level variables. As more genetic factors are identified that might account for treatment variability between individuals, IPDMAs offer a powerful strategy that can be used on existing trial data sets. Despite practical difficulties, IPDMAs are increasingly being conducted.

Clinical Trial Results Applied to Management of the Individual Cancer Patient

The application of clinical trial results to the management of the individual cancer patient is not always straightforward. The results of a clinical trial indicate the "average" effect of an intervention, often expressed in terms of an absolute risk reduction, which is an estimate of the likelihood of benefit for a particular patient. However, within any clinical trial, there might be differences between groups of patients in underlying pathology, genetics, or biology, and some patients might benefit more from a new treatment than others. Thus, within a clinical trial, it might also be useful to group together patients with similar characteristics, and test for subgroup interaction. The test for interaction will indicate whether the magnitude of benefit differs from one prognostic subgroup to the next (a quantitative interaction). Much less common are qualitative interactions, in which a new treatment is beneficial in one subgroup but harmful in another. If the test for subgroup interaction is significant, then the effects of treatment may indeed differ between subgroups of patients, but this should be confirmed in other trials before a treatment is implemented in clinical practice.

Heart failure, chronic diuretic use, and increase in mortality and hospitalization: an observational study using propensity score methods

AIMS

Non-potassium-sparing diuretics are commonly used in heart failure (HF). They activate the neurohormonal system, and are potentially harmful. Yet, the long-term effects of chronic diuretic use in HF are largely unknown. We retrospectively analysed the Digitalis Investigation Group (DIG) data to determine the effects of diuretics on HF outcomes.

METHODS AND RESULTS

Propensity scores for diuretic use were calculated for each of the 7788 DIG participants using a non-parsimonious multivariable logistic regression model, and were used to match 1391 (81%) no-diuretic patients with 1391 diuretic patients. Effects of diuretics on mortality and hospitalization at 40 months of median follow-up were assessed using matched Cox regression models. All-cause mortality was 21% for no-diuretic patients and 29% for diuretic patients [hazard ratio (HR) 1.31; 95% confidence interval (CI) 1.11-1.55; P = 0.002]. HF hospitalizations occurred in 18% of no-diuretic patients and 23% of diuretic patients (HR 1.37; 95% CI 1.13-1.65; P = 0.001).

CONCLUSION

Chronic diuretic use was associated with increased long-term mortality and hospitalizations in a wide spectrum of ambulatory chronic systolic and diastolic HF patients. The findings of the current study challenge the wisdom of routine chronic use of diuretics in HF patients who are asymptomatic or minimally symptomatic without fluid retention, and are on complete neurohormonal blockade. These findings, based on a non-randomized design, need to be further studied in randomized trials.

Using the ROC curve for gauging treatment effect in clinical trials

Non-parametric procedures such as the Wilcoxon rank-sum test, or equivalently the Mann-Whitney test, are often used to analyse data from clinical trials. These procedures enable testing for treatment effect, but traditionally do not account for covariates. We adapt recently developed methods for receiver operating characteristic (ROC) curve regression analysis to extend the Mann-Whitney test to accommodate covariate adjustment and evaluation of effect modification. Our approach naturally extends use of the Mann-Whitney statistic in a fashion that is analogous to how linear models extend the t-test. We illustrate the methodology with data from clinical trials of a therapy for Cystic Fibrosis.

Gender-associated risk factors for cardiac end points and total mortality after renal transplantation: post hoc analysis of the ALERT study

BACKGROUND

Female gender offers a cardioprotective effect over men in the general population, but is lost in the dialysis population. Whether renal transplantation restores the gender-dependent cardiac protection and whether there is a difference in the impact of risk factors is not known.

METHODS

This is a post hoc analysis of pre-defined end points in the placebo arm in the Assessment of Lescol in Renal Transplantation (ALERT) study, a study in renal transplant recipients. Cox regression was performed to estimate the association between different risk factors at baseline and non-fatal myocardial infarction (MI) or cardiac death and total mortality, and specifically assess whether there are gender differences.

RESULTS

The placebo arm included 1052 patients (mean age 50.1 +/- 11.1 yr, 65.3% males) with a mean follow-up of 65 months. The incidence of non-fatal MI or cardiac death was 10.9% vs. 7.9% (NS) and total mortality 13.3% vs. 12.8% (NS) in men and women. In multivariate analysis, previous coronary heart disease (CHD), diabetes, treatment for rejection and serum triglycerides were predictive for cardiac events in men, and low-density lipoprotein (LDL)/high-density lipoprotein (HDL) ratio only in women. A slightly different risk-factor pattern appeared for total mortality. Diabetes, ECG abnormalities, plasma triglycerides, serum creatinine, time on dialysis and age predicted total mortality in men, while ECG abnormalities, LDL/HDL ratio and age were predictors in women.

CONCLUSION

In this relatively low-risk population of renal transplant recipients, no gender difference in cardiac events or total mortality was observed, suggesting that female gender advantage regarding CHD is not restored following transplantation. The predictive value of risk factors differed in men and women.

Is There Evidence to Support the Need for Routine Surgeon Presence on Trauma Patient Arrival?

The trauma center certification requirements of the American College of Surgeons include the expectation that, whenever possible, general surgeons be routinely present at the emergency department arrival of seriously injured patients. The 2 historical factors that originally prompted this requirement, frequent exploratory laparotomies and emergency physicians without trauma training, no longer exist in most modern trauma centers. Research from multiple centers and in multiple varying formats has not identified improvement in patient-oriented outcomes from early surgeon involvement. Surgeons are not routinely present during the resuscitative phase of Canadian and European trauma care, with no demonstrated or perceived decrease in the quality of care. American trauma surgeons themselves do not consistently believe that their use in this capacity is either necessary or an efficient distribution of resources. There is not compelling evidence to support the assumption that trauma outcomes are improved by the routine presence of surgeons on patient arrival. Research is necessary to clarify which trauma patients require either emergency or urgent unique expertise of a general surgeon during the initial phase of trauma management. Individual trauma centers should be permitted the flexibility necessary to perform such research and to use such findings to refine and focus their secondary triage criteria.

Multivariable risk prediction can greatly enhance the statistical power of clinical trial subgroup analysis

Background

When subgroup analyses of a positive clinical trial are unrevealing, such findings are commonly used to argue that the treatment's benefits apply to the entire study population; however, such analyses are often limited by poor statistical power. Multivariable risk-stratified analysis has been proposed as an important advance in investigating heterogeneity in treatment benefits, yet no one has conducted a systematic statistical examination of circumstances influencing the relative merits of this approach vs. conventional subgroup analysis.

Methods

Using simulated clinical trials in which the probability of outcomes in individual patients was stochastically determined by the presence of risk factors and the effects of treatment, we examined the relative merits of a conventional vs. a "risk-stratified" subgroup analysis under a variety of circumstances in which there is a small amount of uniformly distributed treatment-related harm. The statistical power to detect treatment-effect heterogeneity was calculated for risk-stratified and conventional subgroup analysis while varying: 1) the number, prevalence and odds ratios of individual risk factors for risk in the absence of treatment, 2) the predictiveness of the multivariable risk model (including the accuracy of its weights), 3) the degree of treatment-related harm, and 5) the average untreated risk of the study population.

Results

Conventional subgroup analysis (in which single patient attributes are evaluated "one-at-a-time") had at best moderate statistical power (30% to 45%) to detect variation in a treatment's net relative risk reduction resulting from treatment-related harm, even under optimal circumstances (overall statistical power of the study was good and treatment-effect heterogeneity was evaluated across a major risk factor [OR = 3]). In some instances a multi-variable risk-stratified approach also had low to moderate statistical power (especially when the multivariable risk prediction tool had low discrimination). However, a multivariable risk-stratified approach can have excellent statistical power to detect heterogeneity in net treatment benefit under a wide variety of circumstances, instances under which conventional subgroup analysis has poor statistical power.

Conclusion

These results suggest that under many likely scenarios, a multivariable risk-stratified approach will have substantially greater statistical power than conventional subgroup analysis for detecting heterogeneity in treatment benefits and safety related to previously unidentified treatment-related harm. Subgroup analyses must always be well-justified and interpreted with care, and conventional subgroup analyses can be useful under some circumstances; however, clinical trial reporting should include a multivariable risk-stratified analysis when an adequate externally-developed risk prediction tool is available.

Serum digoxin concentration and outcomes in women with heart failure: A bi-directional effect and a possible effect modification by ejection fraction

BACKGROUND

The association between serum digoxin concentration (SDC) and outcomes in women with heart failure (HF) has not been well studied.

AIMS

To test the hypothesis that the effect of digoxin on outcomes in women with HF is bi-directional and dependent on SDC, as in men, and is modified by ejection fraction (EF).

METHODS

We studied 1366 female participants of the Digitalis Investigation Group trial in whom data on SDC (ng/ml) were available. We calculated adjusted odds ratios (AOR) and Bonferroni-adjusted 97.5% confidence intervals (CI) for various outcomes at a median follow up of 41 months, in all women and stratified by EF 35%.

RESULTS

Compared with placebo (26.9%), 40.3% with SDC> or =1.2 (AOR=1.80; CI=1.14-2.86; p=0.004) and 26.6% with SDC 0.5-1.1 (AOR=1.05; CI=0.73-1.51; p=0.762) died. Respective rates for HF-hospitalizations were: placebo (32.8%), SDC> or =1.2 (38.0%) and SDC 0.5-1.1 (25.5%). For women with EF<35% (N=677), SDC 0.5-1.1 lowered odds for HF-hospitalizations (AOR=0.63; CI=0.39-1.00; p=0.026) without increasing odds for death (AOR=0.77; CI=0.47-1.26; p=0.233). In women with EF> or =35% (N=689), SDC 0.5-1.1 had a borderline association with death (AOR=1.58; CI=0.92-2.72; p=0.058) but not with HF-hospitalization (AOR=0.95; CI=0.54-1.66; p=0.826).

CONCLUSIONS

As in men, in women with HF, digoxin has a bi-directional effect based on SDC, and the beneficial effects were significant only among women with EF<35%.

Are all dilemmas in gerontology being swept under the carpet of intra-individual variability?

It is considered that there are great differences among elderly individuals, because the intra-individual variability is great. The differences among specific individuals grow with their age, so when adults reach a very old age, it seems that there are great differences among them--some are able to do some work, the others are not so able-bodied, whereas among high school students there is usually little difference in their physical ability. The research that supports the above mentioned points, however, does not exist and this opinion came about as a result of deduction. The goal of this study is to examine the fluctuations in the elderly and prove that the genetic difference plays a bigger role than the variability, as the intra-individual (or the between-person) variability is present everywhere, not only in very old people.

Tumor-specific expression of vascular endothelial growth factor receptor 2 but not vascular endothelial growth factor or human epidermal growth factor receptor 2 is associated with impaired response to adjuvant tamoxifen in premenopausal breast cancer

PURPOSE

Vascular endothelial growth factor A (VEGF-A) and vascular endothelial growth factor receptor 2 (VEGFR2) are often coexpressed in breast cancer, and potentially affect cellular pathways and key proteins such as the estrogen receptor (ER) targeted by endocrine treatment. We therefore explored the association between adjuvant tamoxifen treatment in breast cancer and expression of VEGF-A and VEGFR2, as well as human epidermal growth factor receptor 2 (HER2), which represents a candidate gene product involved in tamoxifen resistance.

PATIENTS AND METHODS

Immunohistochemical expression of tumor-specific VEGF-A, VEGFR2, and HER2 was evaluated in tumor specimens from premenopausal breast cancer patients randomly assigned to 2 years of tamoxifen or no treatment (n = 564), with 14 years of follow-up. Hormone receptor status was determined in 96% of the tumors.

RESULTS

VEGF-A, VEGFR2, and HER2 were assessable in 460, 472, and 428 of the tumors, respectively. In patients with ER-positive and VEGFR2-low tumors, adjuvant tamoxifen significantly increased recurrence-free survival (RFS; [HR] hazard ratio for RFS, 0.53; P = .001). In contrast, tamoxifen treatment had no effect in patients with VEGFR2-high tumors (HR for RFS, 2.44; P = .2). When multivariate interaction analyses were used, this difference in treatment efficacy relative to VEGFR2 expression status was statistically significant for both ER-positive (P = .04) plus ER-positive and progesterone receptor-positive tumors. We found no significant difference in tamoxifen treatment effects in relation to VEGF-A or HER2 status.

CONCLUSION

Tumor-specific expression of VEGFR2 was associated with an impaired tamoxifen effect in hormone receptor-positive premenopausal breast cancer. Tamoxifen in combination with VEGFR2 inhibitors might be a novel treatment approach for VEGFR2-expressing breast cancer, and such a treatment might restore the tamoxifen response.

Intermediate versus hard end points in clinical trials on hypertension

The main features of randomized clinical trials performed in the field of hypertension treatment are reviewed, based either on the evaluation of clinical events or on the assessment of surrogate outcome measures. The advantages, limitations, and clinical relevance of each approach are highlighted.

Contribution of primary care to health systems and health

Evidence of the health-promoting influence of primary care has been accumulating ever since researchers have been able to distinguish primary care from other aspects of the health services delivery system. This evidence shows that primary care helps prevent illness and death, regardless of whether the care is characterized by supply of primary care physicians, a relationship with a source of primary care, or the receipt of important features of primary care. The evidence also shows that primary care (in contrast to specialty care) is associated with a more equitable distribution of health in populations, a finding that holds in both cross-national and within-national studies. The means by which primary care improves health have been identified, thus suggesting ways to improve overall health and reduce differences in health across major population subgroups.