Concepts of Clinical Trials

Transcription

1 Concepts of Clinical Trials Samiran Sinha Texas A&M University November 18, 2017 Samiran Sinha (TAMU) Clinical trial November 18, / 56

2 What are clinical trials and why do people participate? Clinical trials are methods (or studies) for investigating safety, side-effects, and effectiveness of a drug or a medical device. Clinical trials are also used to compare several drugs. Healthy and unhealthy people (ill) participate in clinical trials. Healthy people do it to help the medical research or to get some financial benefit. People with condition (disease) take part in clinical trials with the hope that they help the science, possibly get a newer medicine, to improve the quality of their life, or get some financial benefit Motives for participating in a clinical research trial: a pilot study in Brazil by Nappo et al. appeared in BMC Public Health Samiran Sinha (TAMU) Clinical trial November 18, / 56

3 Some concepts Before a clinical trial, usually the therapy (treatment) is tested in laboratories, and tested on animals. Using trials on human subjects we gain more understanding about the therapy, how it works, its side effect, the effectiveness of the therapy etc. Protocol: A clinical trial is carried out according to a plan known as a protocol. The protocol is carefully designed to safeguard the participants health and answer specific research questions. A protocol includes the following aspect of the trial design: a) trial purpose, b) who are eligible to participate in the trial and how many subjects to be included, c) details about tests, procedures, medications, and dosages, d) the length of the study and what information will be gathered, e) how the subjects to be followed, f) when to stop the experiment, g) informed consent form Samiran Sinha (TAMU) Clinical trial November 18, / 56

4 Some concepts (continues) A clinical study is usually led by a principal investigator (PI), who is often a doctor. Members of the research team regularly monitor the participants health to determine the study s safety and effectiveness. IRB review: Before the study is conducted, this study protocol is reviewed by Institutional Review Board (IRB). This review board ensures, safety, welfare, and confidentiality of the human subjects. IRB also makes sure that ethical issues are maintained in a trial. Samiran Sinha (TAMU) Clinical trial November 18, / 56

5 Some concepts (continues) Sponsors: Clinical trials are sponsored by the government agencies, pharmaceutical companies, other corporate organizations. Informed consent: Informed consent is a process of informing potential participants about the potential risk and potential benefit of the trials. This helps subjects to decide to take part in the study. Samiran Sinha (TAMU) Clinical trial November 18, / 56

6 Some classification of clinical trials Two main types: observational and interventional trial (Thiese, 2014) Observational clinical trial: progression of a disease and the potentially influential factors are studied. Usually cohort, case-control, and cross-sectional studies are used to investigate these issues. Interventional trial: prevention trial, diagnostic trial, screening trial, treatment trials, quality of life trials Samiran Sinha (TAMU) Clinical trial November 18, / 56

7 Prevention trials These trials look for better ways to prevent a disease in people who have never had the disease or to prevent the disease from returning. Better approaches may include medicines, vaccines, or lifestyle changes, among other things. In this trial, subjects initially do not have the disease or the condition. One example of a prevention trial is the IBIS 2 breast cancer prevention trial. The goal of this trial was to investigate if the drug anastrozole prevent breast cancer development in post menopausal women who were at a high risk of getting it. The trial results showed that the drug reduce the risk of developing breast cancer in these women. Samiran Sinha (TAMU) Clinical trial November 18, / 56

8 Diagnostic trials These trials test the presence of a disease among the subjects with some symptoms. Here we are more concered about the high specificity of the test (high chance of diagnosing a true negative as negative). Samiran Sinha (TAMU) Clinical trial November 18, / 56

9 Screening trials Screening trials test the best way to detect the presence of certain diseases or health conditions at the early stage. For controlling or for cure, for some diseases it is important to detect their presence before the symptoms show up. For these type of diseases, such as bowel cancer and pancreatic cancer, there have been several screening trials. Samiran Sinha (TAMU) Clinical trial November 18, / 56

10 Quality of life trials Quality of life trials (or supportive care trials) explore and measure ways to improve the comfort and quality of life of people with a chronic illness. Samiran Sinha (TAMU) Clinical trial November 18, / 56

11 Main phases of a clinical trial (for interventional trials) Phase I trials: Researchers test an experimental drug or treatment in a small group of people (10 50) for the first time. The purpose is to find out if a new treatment is safe, to find the best way to give the new treatment, such as by mouth or by vein, to see if there are signs that cancer responds to the new treatment. Phase I clinical trials often last several months to a year. For a new drug trial for cancer treatment, doctors offer treatment in phase I clinical trials to people whose cancers wont respond to standard treatments. Although phase 1 design is not for testing how well a treatment or combination of treatments works, the treatment given during this phase may help to slow or stop the growth of a persons cancer. Samiran Sinha (TAMU) Clinical trial November 18, / 56

12 Phase II The purpose is to test safety of the drug, know more about possible side effects, how effective the treatment is. Usually around 100 patients join a Phase II trial. If the new treatment works, doctors may go on to study it in a Phase III trial. Samiran Sinha (TAMU) Clinical trial November 18, / 56

13 Phase III The treatments that show effectiveness at least under some condition in Phase II trial go to Phase III trial. The experimental drug or treatment is administered to a large group of people ( or more) to confirm its effectiveness, monitor side effects, compare it with standard or equivalent treatments, and collect information that will allow the experimental drug or treatment to be used safely. Samiran Sinha (TAMU) Clinical trial November 18, / 56

14 Phase IV Phase IV trials are conducted after a drug is approved by the FDA and made available to the public. In this trial researchers track its long-term safety, seek more information about a drug or treatments risks, benefits, and optimal use. Usually this trial may involve a lot more people with a diverse background than that for a Phase III trial (Suvarna, 2010). Samiran Sinha (TAMU) Clinical trial November 18, / 56

15 What is randomization? Random assignment of treatments among subjects is called randomization. This is mainly used in Phase II or Phase III trials. Randomization is used to remove the selection bias, and to minimize the effect of uncontrollable confounding variables. There are different kinds of randomization, see Suresh (2011) for details. Samiran Sinha (TAMU) Clinical trial November 18, / 56

16 Concept of blinding Blinding is a procedure to keep people unaware of which treatment a subject is assigned to. This is done to avoid unconscious or conscious bias. A study (trial) can fall in one of the following categories based on blinding. Category un-blinded single blinded double blinded Details All parties know the assigned treatment to a patient Only the patient does not know which treatment he/she is receiving The patient and the clinician/data collector do not know which treatment the patient is receiving Samiran Sinha (TAMU) Clinical trial November 18, / 56

17 Sample size calculation We need the sample size, the number of subjects to be included in the study, for the protocol. The total cost of the trial depends on the sample size. Without enough number of subjects in the study, statistical tests may not detect the effectiveness of the treatment. Samiran Sinha (TAMU) Clinical trial November 18, / 56

18 Things needed for the sample size Type-I error rate (α) and Type-II error rate Hypotheses (null and alternative) The primary end point of the study The expected response under the treatment and control (these expected values must be clinically meaningfully different under the alternative hypothesis) Drop-out rate For details see Sakpal (2010). Samiran Sinha (TAMU) Clinical trial November 18, / 56

19 Protocol deviation A protocol deviation occurs when a patient departs from the defined experimental procedure (either drops out from the study, does not follow the instructions of taking the medicine etc) Here is one such example. A randomized double-blind trial compared a low dose of new antidepressant with a high dose of the drug and with a control treatment. The data are Effectiveness low high control measure very effective effective ineffective total assessed withdrawn Samiran Sinha (TAMU) Clinical trial November 18, / 56

20 Protocol deviation continues So the withdrawn patients do not follow the protocol a violation. Here are the options for analyzing such data: per protocol analysis and intent-to-treat analysis. per protocol analysis: ignore the withdrawn subjects in the analysis intention-to-treat analysis: include the withdrawn subjects and consider their responses as ineffective The results of these two analyses could be very different see the next page Samiran Sinha (TAMU) Clinical trial November 18, / 56

21 Protocol deviation continues : proportions based on per-protocol analysis : proportions based on intent-to-treat analysis Effectiveness low high control measure very effective 0.22(0.13) 0.66(0.4) 0.43(0.40) effective 0.44(0.27) 0.17(0.1) 0.57(0.53) ineffective 0.33(0.60) 0.17(0.5) 0.00(0.07) total assessed withdrawn Samiran Sinha (TAMU) Clinical trial November 18, / 56

22 Interim analysis This is analysis of the trial data before the clinical trial is completed. Thus, here analysis is based on the partial data from the trial. The plan for the interim analysis including the statistical approaches should be disclosed in the protocol. Samiran Sinha (TAMU) Clinical trial November 18, / 56

23 Analysis of two-sample binary data Test to compare two proportions Population 1, success probability π 1, sample from population 1 is (X 1,..., X m ), each X i is a binary variable Population 2, success probability π 2, sample from population 1 is (Y 1,..., Y n ), each Y i is a binary variable The estimator of π 1 and π 2 are π 1 = m i=1 X i/m and π 2 = n i=1 Y i/n The null hypothesis is H 0 : π 1 = π 2, there are three different alternative hypotheses Test statistic SE = T = π 1 π 2, SE π(1 π) ( 1 m + 1 n π = ( m i=1 X i + n i=1 Y i)/(m + n) common mean proportion under H 0 For large m, n, T follows N(0, 1) distribution under H 0 Samiran Sinha (TAMU) Clinical trial November 18, / 56 ),

24 Large sample rejection rule H a Rejection rule for level α H a : π 1 > π 2 Reject if T > Z α H a : π 1 < π 2 Reject if T < Z α H a : π 1 π 2 Reject if T > Z α/2 Z r : pr(z > Z r ) = r for some r Samiran Sinha (TAMU) Clinical trial November 18, / 56

25 Large sample confidence interval H a Confidence interval One-sided {, π 1 π 2+Z α π1(1 π 1)/m + π 2(1 π 2)/n} (upper bound of interest) One-sided { π 1 π 2 Z α π1(1 π 1)/m + π 2(1 π 2)/n, } (lower bound of interest) Two-sided π 1 π 2± Z α/2 π1(1 π 1)/m + π 2(1 π 2)/n }{{} margin of error Samiran Sinha (TAMU) Clinical trial November 18, / 56

26 Example Hypothetical example: Suppose that in a randomized double blinded two parallel arm clinical trial the primary end point is the occurrence of an event over the trial period. Observed event rate are 60/150 and 70/155. Is there statistically significant treatment effect? Samiran Sinha (TAMU) Clinical trial November 18, / 56

27 Sample size calculation The required sample size n for each group if we want to construct a (1 α)100% confidence interval for π 1 π 2 with a margin of error ES is n = {π 1(1 π 1) + π 2(1 π 2)} ( ) 2 Zα/2 The required sample size n for each group if we want to reject H 0 with power (1 β) when π 1 π 2 is ES where ( ) Zα/2 + Z 2 β n = 2, ES ES = π1 π2 π(1 π) with π = (π 1 + π 2)/2 Samiran Sinha (TAMU) Clinical trial November 18, / 56

28 Sample size calculation continues Besides specification of α and β, in the sample size calculation we need to know π 1 and π 2 beforehand. That information often comes from a prior trial. If the drop out rate for the kth group is m k ( (0, 1)), then the required sample size if n/m k. Samiran Sinha (TAMU) Clinical trial November 18, / 56

29 Clinical trial case study volunteers-wanted-for-ptsd-study-of-treatment-some-call-a-miracle Samiran Sinha (TAMU) Clinical trial November 18, / 56

30 Analysis of data from cross-over trial A real example of cross-over trial can be found here https: //link.springer.com/content/pdf/ %2fs y.pdf Samiran Sinha (TAMU) Clinical trial November 18, / 56

31 Real datasets on clinical trial Some clinical trial datasets are available at Before the analysis, you may need to pre-process the data. Samiran Sinha (TAMU) Clinical trial November 18, / 56

32 Cross-over clinical trial with binary response There 2 are two treatments, standard treatment A and experimental treatment B, and two groups, gr 1: A B, and gr 2: B A Y giz : Outcome of patient i (i = 1,..., n g ) in period z (z = 1, 2), in group g (g = 1, 2) Y giz = 1 for success and Y giz = 0 for failure Our model is pr(y giz = 1 Treat) = p gi exp{ηi (Treat = B) + γi (z = 2)}, where η : treatment effect (effect of B with respect to treatment A) γ: period effect (effect of period 2 with respect to period 1) p gi : probability of success when the ith subject in the gth group receives treatment A in period 1. We assume that p gi is random, and p gi f g (p gi ) 2 CSDA, 2012, Vol 56, p Samiran Sinha (TAMU) Clinical trial November 18, / 56

33 Cross-over clinical trial continues Treatment effect Period effect PR T = pr(y giz = 1 Treat = B) pr(y giz = 1 Treat = A) = exp(η) PR P = pr(y gi2 = 1 Treat) pr(y gi1 = 1 Treat) = exp(γ) Our goal is estimation of PR T (or η) and PR P (or γ) Samiran Sinha (TAMU) Clinical trial November 18, / 56

34 Cross-over clinical trial continues The observed data in each group can be classified according to the following tables Group 1 Group 2 Response in period 2 Response in period Response 0 n 100 (π 100 ) n 101 (π 101 ) n 200 (π 200 ) n 201 (π 201 ) in period 1 1 n 110 (π 110 ) n 101 (π 101 ) n 200 (π 200 ) n 201 (π 201 ) Define n 1++ = n n n n 111, n 2++ = n n n n 211 π grc = pr(randomly chosen patient from group g exhibits response r and c in period 1 and 2 respectively), g = 1, 2, r, c = 0, 1 Samiran Sinha (TAMU) Clinical trial November 18, / 56

35 Cross-over clinical trial continues The observed data in each group can be classified according to the following tables Group 1 Group 2 Response in period 2 Response in period Response 0 n 100 (π 100 ) n 101 (π 101 ) n 200 (π 200 ) n 201 (π 201 ) in period 1 1 n 110 (π 110 ) n 101 (π 101 ) n 200 (π 200 ) n 201 (π 201 ) Define n 1++ = n n n n 111, n 2++ = n n n n 211 π grc = pr(randomly chosen patient from group g exhibits response r and c in period 1 and 2 respectively), g = 1, 2, r, c = 0, 1 Samiran Sinha (TAMU) Clinical trial November 18, / 56

36 Cross-over clinical trial continues π grc can be estimated by π grc = n grc /n g++ However, we can express π grc in terms of η and γ, so eventually we should be able to estimate η and γ. Samiran Sinha (TAMU) Clinical trial November 18, / 56

37 Cross-over clinical trial continues Note that π 111 = π 110 = π 101 = π 100 = p 1i p 1i exp(η + γ)f 1 (p 1i )dp 1i, p 1i {1 p 1i exp(η + γ)}f 1 (p 1i )dp 1i, (1 p 1i )p 1i exp(η + γ)f 1 (p 1i )dp 1i, (1 p 1i ){1 p 1i exp(η + γ)}f 1 (p 1i )dp 1i Samiran Sinha (TAMU) Clinical trial November 18, / 56

38 Cross-over clinical trial continues Next π 1+1 = π π 111 = pr(randomly chosen subject in group 1 has positive response in period 2) = pr(randomly chosen subject in group 1 has positive response due to treat. B) = exp(η + γ)µ 1, where µ 1 = p 1i f 1 (p 1i )dp 1i Samiran Sinha (TAMU) Clinical trial November 18, / 56

39 Cross-over clinical trial continues Next π 11+ = π π 111 = pr(randomly chosen subject in group 1 has positive response in period 1) = pr(randomly chosen subject in group 1 has positive response due to treat. A) = p 1i {1 p 1i exp(η + γ)}f 1 (p 1i )dp 1i + p 1i p 1i exp(η + γ)f 1 (p 1i )dp 1i = µ 1 Therefore, θ 1 = π 1+1 /π 11+ = exp(η + γ) Samiran Sinha (TAMU) Clinical trial November 18, / 56

40 Cross-over clinical trial continues Note that π 211 = π 210 = π 201 = π 200 = p 2i exp(η)p 2i exp(γ)f 2 (p 2i )dp 2i, p 2i exp(η){1 p 2i exp(γ)}f 2 (p 2i )dp 2i, {1 p 2i exp(η)}p 2i exp(γ)f 2 (p 2i )dp 2i, {1 p 2i exp(η)}{1 p 2i exp(γ)}f 2 (p 2i )dp 2i Samiran Sinha (TAMU) Clinical trial November 18, / 56

41 Cross-over clinical trial continues Note that π 211 = π 210 = π 201 = π 200 = p 2i exp(η)p 2i exp(γ)f 2 (p 2i )dp 2i, p 2i exp(η){1 p 2i exp(γ)}f 2 (p 2i )dp 2i, {1 p 2i exp(η)}p 2i exp(γ)f 2 (p 2i )dp 2i, {1 p 2i exp(η)}{1 p 2i exp(γ)}f 2 (p 2i )dp 2i Samiran Sinha (TAMU) Clinical trial November 18, / 56

42 Cross-over clinical trial continues Next π 21+ = π π 211 = pr(randomly chosen subject in group 2 has positive response in period 1) = pr(randomly chosen subject in group 2 has positive response due to treat. B) = p 2i exp(η)p 2i exp(γ)f 2 (p 2i )dp 2i + p 2i exp(η){1 p 2i exp(γ)}f 2 (p 2i )dp 2i = exp(η)µ 2, where µ 2 = p 2i f 2 (p 2i )dp 2i Samiran Sinha (TAMU) Clinical trial November 18, / 56

43 Cross-over clinical trial continues Next π 2+1 = π π 211 = pr(randomly chosen subject in group 2 has positive response in period 2) = pr(randomly chosen subject in group 2 has positive response due to treat. A) = {1 p 2i exp(η)}p 2i exp(γ)f 2 (p 2i )dp 2i, + p 2i exp(η)p 2i exp(γ)f 2 (p 2i )dp 2i, = exp(γ)µ 2 Therefore, θ 2 = π 2+1 /π 21+ = exp(γ η). Combining these results we get or θ 1 exp(η + γ) = θ 2 exp(γ η) = exp(2η) exp(η) = θ1 θ 2 Samiran Sinha (TAMU) Clinical trial November 18, / 56

44 Cross-over clinical trial continues θ 1 can be estimated by θ 2 can be estimated by θ 1 = π π 111 π π 111 θ 2 = π π 211 π π 211 êxp(η) = θ1 θ 2 Samiran Sinha (TAMU) Clinical trial November 18, / 56

45 Cross-over clinical trial continues Note that Var[log{êxp(η)}] 1 π110 + π 101 4{ n 1 ( π 11+ π 1+1 ) + π } π 201 n 2 ( π 21+ π 2+1 ) The approximate 95% CI for exp(η) is exp((1/2)log( θ 1 / θ 2 ) ± 1.96 Var[log{êxp(η)}]) Samiran Sinha (TAMU) Clinical trial November 18, / 56

46 Cross-over clinical trial continues Note that θ 1 θ 2 = exp(η + γ) exp(γ η) = exp(2γ) Therefore, exp(γ) can be estimated by θ 1 θ2 The asymptotic variance of log{êxp(γ)} is { 1 π110 + π n 1 ( π 11+ π 1+1 ) + π } π 201 n 2 ( π 21+ π 2+1 ) Yes, it is the same variance of log{êxp(η)} Samiran Sinha (TAMU) Clinical trial November 18, / 56

47 Cross-over clinical trial continues To check the presence of interaction (carry-over efect) we take the model pr(y giz = 1 Treat) = p gi exp{ηi (Treat = B) + γi (z = 2) + δi (Treat = B)I (z = 2)}, If δ = 0 there is no carry-over effect Here π 111 = p 1i p 1i exp(η + γ + δ)f 1 (p 1i )dp 1i = exp(η + γ + δ) p1if 2 1 (p 1i )dp 1i π 211 = p 2i exp(η)p 2i exp(γ)f 2 (p 2i )dp 2i = exp(η + γ) p2if 2 2 (p 2i )dp 2i Under random assignment of the subjects into the two groups, f 1 ( ) = f 2 ( ), consequently, p 2 1i f 1(p 1i )dp 1i = p 2 2i f 2(p 2i )dp 2i Samiran Sinha (TAMU) Clinical trial November 18, / 56

48 Cross-over clinical trial continues Therefore, exp(δ) = π 111 π 211 We can estimate δ by δ = log( π 111 ) log( π 211 ) The variance of δ is Var{log( π 111 )} + Var{log( π 211 )} that can be estimated by τ 2 = (1 π 111) + (1 π 211) n 1 π 111 n 2 π 211 An asymptotic (1 α)100% CI for δ δ ± Z α/2 τ Samiran Sinha (TAMU) Clinical trial November 18, / 56

49 Cross-over clinical trial continues 3 A two-period double blind crossover trial of 12 µg formoterol solution compared with 200 µg salbutamol solution administered to 24 children with exercise induced athsma. Response is coded as S and F corresponding to good and not good based upon the investigators overall assessment. Subjects were randomized to one of two groups: group 1 received the treatments in the order formoterol salbutamol; group 2 in the order salbutamol formoterol. The results are given below: 3 Cross-over Trials in Clinical Research by Stephen Senn, 2002 Samiran Sinha (TAMU) Clinical trial November 18, / 56

50 Cross-over clinical trial continues Group 1 Group 2 Subject Formoterol Salbutamol Subject Salbutamol Formoterol 1 S S 13 F S 2 F F 14 F S 3 S F 15 S S 4 S F 16 S S 5 S S 17 S S 6 S F 18 S S 7 S F 19 S S 8 S F 20 S F 9 S F 21 F S 10 S F 22 F S 11 S F 23 F S 12 S F 24 F S Samiran Sinha (TAMU) Clinical trial November 18, / 56

51 Cross-over clinical trial continues The data can be summarized as follows: Group 1 Group Since some cell frequencies are zero, we make continuity correction by adding 0.5 with each cell and obtain Group 1 Group Next test for the carry-over effect. Here δ = log(2.5/5.5) = 0.79 and se( δ) = Hence, the approximate 95% confidence interval for δ is ( 0.79 ± ) includes 0. Therefore, we do not have sufficient evidence to reject H 0 : δ = 0 against H a : δ 0 at the 5% level of significance. That means there is no reason to believe that there is a carry-over effect. Samiran Sinha (TAMU) Clinical trial November 18, / 56

52 Cross-over clinical trial continues Next, you test for the treatment effect and period effect. The potential issues are 1) small sample size and 2) for some subjects we may not have two responses due to drop-out from the study. Samiran Sinha (TAMU) Clinical trial November 18, / 56

53 Cross-over clinical trial continues 4 Here we have measurements of peak expiratory flow (PEF), a measure of lung function, made on 13 children aged 7 to 14 with moderate or severe asthma in a two-period cross-over trial comparing the effects of a single inhaled dose of 200 µg salbutamo and 12 µg formeterol. Here is the dataset. 4 From Chapter 3 of Stephen Senn s book referred earlier Samiran Sinha (TAMU) Clinical trial November 18, / 56

54 Cross-over clinical trial continues Group 1 Group 2 Patient id Formoterol Salbutamol Subject Salbutamol Formoterol Samiran Sinha (TAMU) Clinical trial November 18, / 56

55 Cross-over clinical trial continues We discussed large sample inference procedure, however there are methods for small sample inference (Gart, 1969). Samiran Sinha (TAMU) Clinical trial November 18, / 56

56 References Fieller, N. (1996). Medical Statistics: Clinical Trials. Gart, JJ. (1969). An exact test for comparing matched proportions in crossover designs. Biometrika, 56, Lui, K-J and Chang, K-C. (2012). Estimation of the proportion ratio under a simple crossover trial. Computational Statistics & Data Analysis, 56, Sakpal, T. V. (2010). Sample size estimation in clinical trial. Perspectives in Clinical Research, 1, Suresh, K. P. (2011). An overview of randomization techniques: An unbiased assessment of outcome in clinical research. Journal of Human Reproductive Sciences, 4, Suvarna, V. (2010). Phase IV of drug development. Perspectives in Clinical Research, 1, Thiese, M. (2014). Observational and interventional study design types: An overview. Biochemia Medica, 24, Samiran Sinha (TAMU) Clinical trial November 18, / 56