Donald A. Berry

Interim Analysis in Clinical Trials: The Role of the Likelihood Principle

Abstract In the classical statistical approach, whether or not a clinician looks at accumulating data with the possibility of stopping the trial can affect any inferences drawn from the trial. This is in contrast to the Bayesian and likelihood approaches to statistics. I give examples to illustrate the differences in these approaches and discuss designing clinical trials from a Bayesian point of view.

Pharmacodynamics and side effects of flecainide acetate

We compared side effects with flecainide trough levels and ECG intervals among 43 patients who received flecainide for up to 34 months. Flecainide plasma levels were higher when associated with cardiovascular side effects (mean 1063 ng/ml; range 296 to 2050 ng/ml) than when no side effects occurred (mean 609 ng/ml; range 89 to 1508 ng/ml; P < 0.001). The PR interval (P < 0.001), QRS interval (P < 0.001), and the rate‐corrected QT interval (P < 0.001) were greater at the time of cardiovascular side effects, but the rate‐corrected JT interval was not. The therapeutic‐toxic window for flecainide plasma level was 381 ng/ml (at least 50% probability of efficacy) to 710 ng/ml (<10% probability of cardiovascular side effects). The risk of cardiovascular side effects increases at higher plasma levels of flecainide and is associated with greater increases in the PR and QRS intervals from baseline than are routinely observed during flecainide dosing.

Modified Two-Armed Bandit Strategies for Certain Clinical Trials

Abstract A procedure which maximizes the expected number of successes in a clinical trial involving two treatments can usually be found only by backward induction. Not only is it difficult to find an optimal procedure but, once found, it is difficult to describe and cumbersome to communicate. A procedure is suggested which depends on the information present concerning the treatments. This procedure is easy to calculate and approximates an optimal procedure quite well. It is applicable to trials for which the number of patients is unknown as well as those of known duration.

A comparison of antiarrhythmic drugs for the suppression of ventricular ectopic depolarizations: a meta-analysis.

This article reports the results of a meta-analysis of the effectiveness of antiarrhythmic drugs for the suppression of ventricular ectopic depolarizations. We analyzed 97 published articles that referred to a total of 27 drugs and contained data from 2989 patient-treatment trials; our goal was to determine the number of patients responding to therapy, defined as greater than or equal to 80% suppression of ventricular ectopic depolarizations. By means of logistic regression we tested the effect of 10 clinical and experimental variables on the likelihood of response to therapy. The likelihood of a drug response was significantly affected by the following six variables: increased by the use of dose titration (t = 3.59, p less than 0.0001), increased by the use of a higher daily dose (t = 3.21, p less than 0.0001), decreased by older age (t-2.67, p = 0.004), decreased by the use of blinding (t = -2.28, p = 0.011), increased by treating more male patients (t = 1.72, p = 0.043), and decreased by the presence of cardiovascular disease (t = -1.52, p = 0.064). Incorporating these six variables into our logistic regression model, we adjusted the response rate in each published study and calculated the mean response and standard error for each drug. Of the drugs tested in at least 100 patients, the most effective were amiodarone (estimated response rate 90%), encainide (80%), flecainide (79%), and propafenone (74%). Class IC drugs were significantly more effective than class IB and II drugs (p less than 0.05). With the exception of lorcainide and moricizine, class IC drugs were also more effective than class IA drugs (p less than 0.05). Amiodarone was significantly more effective than all drugs except encainide and flecainide (p less than 0.05). We found no significant differences among the response rates to class IA, IB, and II drugs. Whereas several patient and study characteristics affect the likelihood of response to antiarrhythmic drugs, class IC drugs and amiodarone are significantly more effective than other drugs in suppressing ventricular ectopic depolarizations.

Experimental design for drug development: a bayesian approach

The Bayesian approach to inference and decision making provides an integrated way of addressing the various aspects of drug development, from the early preclinical study of compounds through the clinical and postmarketing phases. In particular, it provides a natural, convenient way for choosing among experimental designs. An essential aspect of the process of evaluating design strategies is the ability to calculate predictive probabilities of potential results. I describe a Bayesian approach to experimental design and illustrate it by considering a particular type of clinical trial. Also, I compare Bayesian and classical statistical attitudes toward design.

Identification of drug interactions by meta-analysis of premarketing trials: The effect of smoking on the pharmacokinetics and dosage requirements for flecainide acetate

Clinical Pharmacology and Therapeutics (1989) 46, 1–8; doi:10.1038/clpt.1989.99

The effects of single-dose atenolol, labetalol, and propranolol on cardiac and vascular function.

The pharmacodynamic effects of single oral doses of atenolol (100 mg), labetalol (300 mg), and propranolol (80 mg) were compared with those of placebo in a randomized, double‐blind, Latin square design in 12 patients with hypertension. Atenolol and propranolol both significantly reduced cardiac output (−0.55 vs. −0.31 L/min) and heart rate (−8.0 vs. −6.6 bpm), whereas labetalol had no effect on either parameter (− 0.08 L/min; +1.0 bpm). Labetalol significantly reduced vascular resistance (− 339 dynes · cm/sec5), but atenolol and propranolol did not (147 vs. 62 dynes · cm/sec5). Only labetalol significantly reduced the systolic (−15.3 mm Hg), diastolic (−11.5 mm Hg), and mean blood pressures (−12.8 mm Hg). Atenolol significantly reduced only diastolic blood pressure (−5.20 mm Hg), whereas propranolol failed to lower these parameters significantly. These data indicate that the hemodynamic profile of labetalol differs from that of selective and nonselective β‐blockers. Labetalol lowered blood pressure primarily by reducing vascular resistance, whereas reductions in heart rate and cardiac output were the predominant effects of atenolol and propranolol.

The pharmacodynamic and pharmacokinetic interaction between single doses of flecainide acetate and verapamil: Effects on cardiac function and drug clearance

Flecainide and verapamil are antiarrhythmic agents that may be used in combination. We have examined their pharmacodynamic interaction by M‐mode echocardiography and electrocardiography in eight normal male volunteers (24 ± 1.8 years of age). Flecainide decreased the left ventricular ejection fraction (LVEF) (−4.4 ± 1.2%, p < 0.008), but verapamil did not. Neither drug affected cardiac output or vascular resistance. Both drugs increased the PR interval (12 ± 4 msec, p < 0.01 for flecainide; 12 ± 5, p < 0.04 for verapamil). Flecainide, but not verapamil, increased the QTc interval (23 ± 8 msec, p < 0.02). Both drugs also increased the systolic time interval ratio (PEPc/LVETc) (0.074 ± 0.012, p < 0.0004 for flecainide; 0.029 ± 0.008, p < 0.007 for verapamil). The combination of flecainide and verapamil had additive effects on myocardial contractility and on atrioventricular conduction. Verapamil slightly decreased the plasma clearance of flecainide (7.78 ± 0.60 ml/kg/min for flecainide alone, 7.34 ± 0.48 ml/kg/min for flecainide and verapamil together, p < 0.05). On the other hand, flecainide had no effect on the plasma clearance of verapamil, which suggests that there was little interaction between the two drugs on their pharmacokinetic parameters.

Statistical inference, designing clinical trials, and pharmaceutical company decisions

The classical design of clinical trials is dictated by the eventual analysis. If the design varies from that planned then classical analysis is impossible. The Bayesian approach on the other hand is completely flexible and is therefore ideal for addressing design questions and practical decision problems. I contrast these two approaches in two types of clinical trials: (i) those that strive to treat patients as effectively as possible and (ii) those sponsored by pharmaceutical companies attempting to maximise their expected profit.

An improved procedure for selecting the better of two bernoulli populations

Abstract A truncated play-the-winner sampling procedure for selecting the better of Bernoulli populations π1 and π2 is considered. Associated with πi is success probability pi where p 1 ≥ p 2. The procedure is designed to select π1 with probability at least P* whenever p 1 − p 2 ≥ Δ *; P* and Δ* are preassigned constants. This truncation procedure is shown to be uniformly asymptotically better than the scheme of Sobel and Weiss [6] with regard to both minimizing the expected total number of trials and minimizing the expected number of trials on π2. Numerical results show that the improvement can be substantial for small values of p 1 and p 2.