Recent Developments in the Analysis of QT Interval Data

Transcription

1 Recent Developments in the Analysis of QT Interval Data Robb Muirhead Statistical Research & Consulting Center Pfizer Global Statistics Graybill Conference VII Colorado State University June 12, Outline Some background on ECGs and QT ICH E14 and Thorough QT Studies Recent work 2

2 Some background on ECGs and QT 3 Electrocardiogram (ECG or EKG) An ECG is a graphic produced by an electrocardiograph, which records the electrical activity of the heart over time The first practical electrocardiograph was invented by Willem Einthoven ( ) in Leiden and used a string galvanometer. The original machine required 5 people to operate it and weighed about 600 lbs. (!!) Einthoven assigned the letters P, Q, R, S, T to various features of an ECG, and described their relation to a number of cardiovascular disorders. Einthoven received the Nobel Prize in Medicine in 1924 for his discovery of the mechanism of the electrocardiogram. 4

3 An Early Electrocardiograph 5 ECG (cont.) An ECG is the gold standard for the diagnosis of cardiac arrhythmias (any of a group of conditions in which the electrical activity of the heart is irregular or is faster or slower than normal) Tachycardia: Heart rate faster than 100 beats/minute Bradycardia: Heart rate slower than 60 beats/minute Fibrillation: Heart muscle begins a quivering motion Atrial: Quivering, chaotic motion in the upper chambers of the heart (not typically a medical emergency) Ventricular: Occurs in the ventricles (lower chambers) leads to cardiac arrest 6

4 ECG of a Healthy Heart Each 1-mm square represents.04 sec (40 msec) in time and.1 mv in voltage. From Lead II (limb lead) About 88 bpm 7 QT Interval P wave reflects atrial depolarization QRS complex reflects ventricular depolarization T wave reflects ventricular repolarization The QT interval is duration of ventricular depolarization and repolarization. QT is the working phase and TQ is the resting phase. RR QT TQ 8

5 Long QT Syndrome and TdP People with the long QT syndrome are susceptible to an abnormally rapid heart rhythm (ventricular tachycardia) called torsades de pointes (TdP), which can lead to sudden death. TdP is characterized by fluctuation of the QRS complexes around the electrocardiographic baseline and is typically associated with a preceding long QT interval. 9 TdP QT prolongation is used as a marker for TdP. The development of a compound may be discontinued if it is associated with QT prolongation. Since 1990, at least 8 drugs (including Seldane and Propulsid) have been withdrawn from major markets because of QT interval prolongation and TdP. (Some of these appear to have increased risk of TdP because drug-drug interactions.) Terfenadine, Astemizole, Grepafloxicin, Terodiline, Lidoflazine, Sertindole, Levomethadyl, Cisapride 10

6 QTc: Corrected QT Because of its inverse relationship to heart rate, the QT interval is routinely transformed (normalized) into a corrected index denoted by QT c. The intent is to represent the QT interval at a standardized heart rate (the QT interval at a heart rate of 60 bmp). Many different formulas have been proposed. The most common have the form QT c =QT/RR a where a is a formula-specific parameter: e.g. a=1/2 gives Bazett s correction a=1/3 gives Fridericia s correction Both of these were published in 1920!!! There are now more than 50 published versions of corrected QT. 11 What s a normal QT c? Generally accepted guidelines: Normal QT c : <430 ms in males, <450 ms in females Borderline QT c : ms (males), ms (females) Prolonged QT c : >450 ms (males), >470 ms (females) The prolongation of the QT interval to longer than 500 msec during drug therapy should prompt a critical reevaluation of the risks and benefits of that therapy. (D. M. Roden, NEJM, 2004) 12

7 One subject from a Pharmacia study Reference regions for 24 hour beat-to-beat data 13 Placebo-corrected QT c : 2 doses of a drug Placebo-corrected QTcF (msec) High dose Low dose 40 mg CP-866, mg CP-866,087 2-sided 90% CIs Time Post Dose (Hr) 14

8 ICH E14 and Thorough QT Studies 15 ICH Guideline (E14) In May, 2005, ICH released a guideline (E14): The Clinical Evaluation of QT/QTc Interval Prolongation and Proarrhythmic Potential for Non-Antiarrhyhmic Drugs. (Available at Section on a thorough QT/QTc study : Drugs are expected to receive a clinical electrocardiographic evaluation typically including a single trial dedicated to evaluating their effect on cardiac repolarization as detected by QT/QTc prolongation. The study is typically carried out in healthy volunteers (as opposed to individuals at increased risk of arrhythmias) and is used to determine whether or not the effect of a drug on the QT/QTc interval in target patient populations should be studied intensively during later stages of drug development. 16

9 ICH E14 There is a single sentence devoted to statistical analysis: E14 says: a negative thorough QT/QTc study is one in which the upper bound of the 95% one-sided confidence interval for the largest time-matched mean effect of the drug on the QTc interval excludes 10 ms. The only reasonable interpretation of this statement is that the CI refers to the maximum of a set of parameters, namely the maximum difference between two sets of population means (one for drug, one for placebo). 17 Simple Model Suppose we have nd subjects on drug and np subjects on placebo, and QT (or QT c ) is measured at k time points. Model: Let X X ( D) 1 ( P) 1,, X,, X ( D) n D ( P) n P ( D) ( P) ~ iid N ~ iid N We want a CI for max. 1 i k i. k k ( D), ( P), 18

10 Simple Model (cont) There s just one minor detail. No satisfactory method has yet been found for constructing such a CI that works under all scenarios 19 CI for (cont.) One large-sample method was investigated in: Eaton ML, Muirhead RJ, Mancuso JY & Kolluri S (2006). A confidence interval for the maximum mean QT interval change due to drug effect. Drug Information Journal, 40, This method doesn t work well if is close to other coordinates of and it doesn t give 1-sided intervals with the right coverage probability. The paper sounds a note of caution. Finding a CI for that is always reliable is an unsolved problem. So what do statisticians do when analyzing data from a thorough QT study? Basically, they re ignoring what ICH E14 says. They have to 20

11 Placebo-corrected QT c : 2 doses of a drug Placebo-corrected QTcF (msec) High dose Low dose 40 mg CP-866, mg CP-866, Time Post Dose (Hr) 21 Hypothesis Testing Approach At each time point, calculate 1-sided 95% confidence intervals for the population mean differences and conclude that the study is negative (drug is non-inferior to placebo) if all the upper bounds are less than or equal to 10 ms. This procedure corresponds to a well-defined multivariate hypothesis testing problem that of testing H H 0 A : i : i for at least one i for all i 1,, k The test described above is both an IUT and the LRT. 22

12 Recent Work 23 Moving beyond thorough QT studies Not all drugs that prolong QT are arrhythmogenic. There is widespread recognition that QT/QTc is an imperfect and/or inadequate biomarker. The QT interval is used during drug development as a surrogate marker for the prediction of a serious adverse drug effect, syncope, or death due to TdP. However, as with many surrogate markers, its relationship to the event of interest is imperfect the risk of TdP is not a linear function of the QT interval, nor of the extent of QT-interval prolongation during drug therapy. Although abnormal QT-interval morphology might predict an increased risk of TdP, analytic methods for assessing the relationship remain to be validated. D.M. Roden, NEJM,

13 Recent Study We need to have a better understanding of issues like: What beat-to-beat variability is to be expected in parameters measured via an ECG? How do we characterize a multivariate region that contains most (e.g. 95%) of ECG parameters in beat-to-beat data over a specified time period (e.g., 1 day)? How reproducible are such regions? A signal of a potential would be represented by movement of beats outside this region. How do we look for this? How comparable is Holter beat-to-beat data to that obtained using traditional ECGs. Pfizer recently carried out a pilot study in which beat-to-beat data was collected for 2 days on 6 healthy males in a CRU. The study was driven by Anthony Fossa, now at icardiac. 25 Recent Study Motivation/scientific hypotheses: Arrhythmogenic risk is influenced by changes in autonomic state. The standard practice of measuring QTc is not appropriate in these situations due to hysteresis of the QT-RR interval relationship. (During rapid changes of the autonomic state, the QT interval does not adapt as quickly as the heart rate (RR), resulting in hysteresis. This can result in a misinterpretation of the risk associated with the QTc value.) The study was intended to be a step toward the development of new quantitative measures to differentiate autonomic responses in normal subjects which might allow for future comparison to subjects with impaired repolarization and risk of arrhythmia. 26

14 Recent Study Four autonomic challenges were given to each subject on Day 1, and repeated on Day 2, and were meant to represent a range of rapid autonomic challenges and subsequent cardiovascular response. The intent was to try to mimic the types of challenges/responses a person would experience over the course of a day. These were: Schellong test (subject stood quickly, remained standing for 1 minute and continued standing for an additional 4 minute recovery) Stationary bicycle for 1 minute at maximal effort, with 200 Watt resistance Isoproterenol to induce tachycardia (speed up HR) Phenylephrine to induce bradycardia (slow down HR) 27 Recent Study Continuous 12-lead Holter recordings were collected for 2 days 28

15 Recent Study Triplicate 12-lead ECGs (separated by approx 2 minutes) were collected on Days 1 and 2 at time 0 and then.5, 1, 2, 2.5, 3, 3.5, 4, 8, 12, 24 hours following time 0. One reason for both Holter and ECG data was to investigate how closely they matched. (Can Holter data be used instead of ECG data in future studies?) The primary reason for the same regimen on both days was to see how reproducible the results were. 29 Recent Study Protocols for Pilot Study (Days 1 and 2) Autonomic Challenges 0Hr 2Hr 3Hr 4Hr Remainder 20 Hr collection Eating Sleeping Start Holter or Standard 12-lead ECG measures Phenylephrine infusion Isoproterenol Infusion Burst exercise bike Schellong rapid standing 30

16 31 Reference Regions One way of studying the effects of autonomic challenges on the QT-RR relationship is to use the 20- hour (11 am to 7 am) beat-to-beat Holter data to construct a baseline reference region that is, a region that contains (for example) 95% of the beats (the (RR, QT) points from these beats). The point of the reference region is to be able to compare normal beats with those from the challenge periods. A standard type of reference region is elliptical, and motivated via the assumption of bivariate normality. This seems to work reasonably well for short time periods or periods of similar activity (e.g. sleep), but not well over long periods. 32

17 2D Histogram 1 Subject from Pharmacia study Over the course of a full day the distribution does not look bivariate normal. 150 Counts QT TQ Hypothesized Arrhythmia Triggering Zone 500 Increased arrhythmogenic risk QT Interval (ms) QT prolongation exceeding baseline autonomic influence 0 normal QT-RR boundary for all levels of autonomic tone RR Interval (ms) 34

18 Mixture of Normal Distributions Reference regions constructed for the 20-hour period were obtained using mixtures of bivariate normal distributions for (RR, QT). A k-component mixture of bivariate normal distributions has a density function of the form p, 1, 1,..., k, i 1 where is the bivariate normal density function; 1,..., k k f x p x p k i i i are the mixing probabilities; are the mean vectors; are the covariance matrices 35 3-Component Mixture Model 1 subject from Pharmacia study Red, blue and green lines are 95% contours for each component. One component models points with high variability The two other components model low & high HR Black line is the 95% reference region for 3-component mixture model with mixing probabilities 0.1, 0.3, 0.59 Observed coverage = 96.25% 36

19 Fitted mixture distribution and 95% reference region. "RLM-08a.txt" 37 Back to recent study 3-component mixtures were used in this study. This choice seems to fit the main cluster of points well using 2 of the 3 components, while the large-variance 3 rd component essentially accounts for the small number of widely varying points (outliers). The mixture distributions were fit by maximum likelihood, using the EM algorithm. 38

20 95% and 99% reference regions: 1 subject Patient Day 1: 20-hour Reference Region (with 2000 random beats) Patient Day 2: 20-hour Reference Region (with 2000 random beats) QT % area: % area: % coverage: 95.69% 99% coverage: 98.61% RR RR QT % area: % area: % coverage: 95.86% 99% coverage: 98.82% RR 600 RR Reproducibility of reference regions How similar are e.g. the 95% reference regions on days 1 and 2? One possible measure of discordance is the percentage of points lying inside one region and outside the other region. One subject had 5.3% discordance; the other 5 were all less than 3.5% 40

21 20-hr reference region discordance Patient Subject : 1: 20-hour Reference Region Comparison Subject 2: Patient : 20-hour Reference Region Comparison QT Day 1 Day 2 Restricted RR Region 5.281% overlap discordance % restricted overlap discordance QT Day 1 Day 2 Restricted RR Region % overlap discordance % restricted overlap discordance RR RR 41 Single subject Patient Day 1: ISO Challenge Patient Day 1: PE Challenge hour Reference Region Before minimum smoothed median After minimum smoothed median 95% total ISO coverage: 80.23% 67.16% outside 20hr RR (green); 90% area = % outside 20hr RR (red); 90% area = hour Reference Region First 200 beats of challenge 95% total PE coverage: 90.91% 26.37% of first 200 beats outside 20hr RR; 90% area = Patient Day 1 Patient Day hour Reference Region 7-9 AM Reference Region 3-5 AM Reference Region 48.07% discordance, 7-9 AM - 20h; 95% area = % discordance, 3-5 AM - 20h; 95% area = % discordance, 7-9 AM AM hour Reference Region 9:00-9:05 AM Schellong Region 9:30-9:35 AM Exercise Region 2.57% Schellong outside 20hr RR; 90% area = % Exercise outside 20hr RR; 90% area = delete these 400 guides from 1000slide 1200 master before 400printing or 1000 giving 1200to 1400 the 1600 client 42

22 Recent Study: A few broad conclusions Holter beat-to-beat data collection and analysis is technically feasible. Holter beat-to-beat values are comparable to standard 12-lead ECG measures during resting state. (These measures are not possible during autonomic change.) There was good reproducibility between days 1 and 2. Assessment of dynamics is possible under conditions of intense autonomic change (both bradycardia and tachycardia) 43 The Future? Subjects in thorough QT studies are in clinics. We could collect beat-to-beat data for the entire period, both on drug and on placebo, for marginal additional expense. There is a larger methodology study in the early planning stage. (Pfizer has approved partial funding for a new study, provided other pharma companies participate.) 44

23 Some References Eaton ML, Muirhead RJ, Mancuso JY & Kolluri S (2006).A confidence interval for the maximum mean QT interval change due to drug effect. Drug Information Journal, 40, Fossa A, Couderc J-P, Martell B, Sieklucki J, Muirhead RJ, Emerson, JW, Dombi T, Zhou M, Trost DC (2008). Quantification of autonomic-mediated changes in dynamic and spatial measures of the beat-to-beat electrocardiogram. Draft manuscript. Roden, DM (2004). Drug-induced prolongation of the QT interval. New England Journal of Medicine, 350,