Article 14 of the Regulations on Clinical Trials of Medical Devices requires that the clinical trial program of medical devices should be tailored to the characteristics of the specific product under test, determine the number of clinical trials, duration and clinical evaluation criteria, so that the results of the trial are statistically significant. Therefore, the statistical issues in the medical device clinical trial is crucial, mainly in the following aspects:
(a) Data management of the trial
In order to ensure the quality of the clinical trial, the sponsor should assign experienced monitors to monitor the whole process of the clinical trial. The monitor serves as a bridge between the investigator and the data manager. In order to ensure the traceability of data, the following should be done:
1. For all subjects in each clinical trial, an original observation record form (e.g., a medical record) and a triplicate, carbonless case report form (CRF) should be established.
2. After each follow-up visit during the course of the trial, the investigator shall fill out the case report form in a timely, accurate, complete, error-free and legible manner. At the end of the trial, the original copy of the case report form should be sent to the data manager by the supervisor, and the remaining two copies should be kept by the investigator and the sponsor respectively, so that timely and effective modifications and corrections can be made to the CRF in case of errors detected by the supervisor of the trial and data management.
3. The data manager should set up a database based on the case report form and ensure that the database runs correctly.
For clinical trials with a long study period and a large number of follow-up visits, in order to identify problems in the implementation of clinical trials and the completion of CRFs at an early stage, and to shorten the time for data management and statistical analysis, it is possible to adopt the practice of sending CRFs to the data management and statistical analysis organization once for each completed follow-up visit.
4, the data manager should also be a preliminary review of each case report form (visual inspection). After the initial review, two computerized data entry personnel independently enter the case report form into the database (two entries), and use the software to compare the results of the two entries (the principle of double-checking). If the data in the two databases do not match, the original case report form should be compared to find out the reason, and scope and logic checks should be carried out.
Only with strict quality control throughout the entire process of the trial can a high-quality database be created to complete the trial program and achieve the original objectives.
5. Clinical statisticians should analyze the data according to the clinical trial protocol and medical record report form, using standard statistical methods and statistical analysis software recognized at home and abroad, and write a statistical analysis report in order to provide the investigator with the basis for writing the clinical trial report.
(II) Calculation of sample size
The purpose of a clinical trial is to collect evidence about the safety and efficacy of a medical device in a sample of the target population, and then to generalize the findings of the trial to the entire population in the real world with the same characteristics as the trial population using statistical analysis. Therefore, a representative sample must be selected for the clinical trial to ensure that scientific and valid conclusions are obtained.
Usually, in order to evaluate the efficacy and safety of the test device, the sample size should be calculated based on the main efficacy evaluation index or safety index, respectively, and the larger one should be used as the sample size of the clinical trial.
However, the sample size calculated based on the safety evaluation indexes is often very large, and it is difficult to implement the clinical trial for the current strength of domestic manufacturers. Therefore, the sample size calculation of the current domestic medical device clinical trials is often based on the main efficacy evaluation index.
The sample size calculation should be based on the purpose of the study to establish the research hypothesis. Research hypotheses are divided into null hypotheses and alternative hypotheses. For example, if the research question is "For a certain disease, after treatment with the test device, is the test device group more effective than the control group"? The two hypotheses for this question are:
1. Null hypothesis H0. The treatment group is less effective than the control group.
2. Alternative hypothesis H1. The efficacy of the treatment group is better than the efficacy of the control group.
The purpose of the implementer and the researcher is to reject the null hypothesis, accept the alternative hypothesis that the treatment group is more effective than the control group, and extrapolate the conclusions drawn from the sample to the population as a whole.
There are two types of decision-making errors that may be made in the process of statistical inference described above, Type I errors (also known as α errors or false-positive errors) and Type II errors (also known as β errors or false-negative errors). We usually call α the significance level and define 1-β as the test efficacy, or degree of certainty.
In general, the size of Type I and Type II errors in clinical trials is clearly defined. Typically, α should not exceed 5% (0.05) and β should not be greater than 20% (with a certainty of not less than 80%).
In the calculation of sample size for hypothesis testing, not only the two error probabilities mentioned above are used, but also the type of test (valid, non-inferiority or equivalence) is taken into account, and non-inferiority or equivalence tests must be performed by specifying the clinically significant difference in efficacy of the treatment group versus the control group, i.e., the difference between the outcome variables determined by a clinical expert to be clinically significant.
In summary, the size of the sample is usually determined by the specific characteristics of the product being tested, the main efficacy indicators and their parameters. The sample size and the basis for its calculation should be included in the clinical trial protocol. Generally speaking, when calculating the sample size, the statistics should be estimated by referring to the published domestic and international literature on the control group, international standards, industry standards, ministerial standards or the results of the pre-test of the product to be tested.
(C) Control of bias
Bias, also known as bias, refers to the systematic errors caused by relevant influencing factors in the design, implementation and statistical analysis of clinical trial protocols and evaluation of the results, making the evaluation of the efficacy or safety of the device deviate from the true value. Bias interferes with the drawing of correct conclusions and must be prevented throughout the entire process of a clinical trial. There are two important control measures:
1. Randomization
In multicenter clinical trials, central randomization should be used to ensure that the treatment and control groups within each study center are balanced and comparable. The randomization table should be generated by a statistician using nationally and internationally available statistical analysis software and be reproducible. The randomization table is a documented arrangement of subjects, i.e., the sequence of treatments.
2. Blinding
Blinding in clinical trials is categorized as double-blind, single-blind, or non-blind (open) depending on the degree of blinding. The degree of blinding required depends on the strength and severity of the potential bias. A single-blind design means that patients do not know whether they are in the treatment or control group; a double-blind design means that neither the patients nor the investigators know which group is the treatment group.
Medical device clinical trials are often unable to be blinded due to ethics, maneuverability, or device specificity, when non-blinded clinical trials can be conducted. However, whether it is a single-blind or non-blind clinical trial, corresponding measures to control trial bias should be developed to minimize possible bias.
(IV) Statistical analysis methods
The statistical analysis methods and statistical analysis software used for data analysis in clinical trials are recognized both at home and abroad. Statistical analyses should be based on correct and complete data, and the correct statistical methods should be selected according to the purpose of the study, the experimental protocol, and the observation indexes. Generally, it can be summarized as the following aspects:
1. Descriptive statistics: generally used for demographic data, baseline data and safety data, including the statistical description of the main indicators and secondary indicators. For example, the mean, standard deviation, maximum value, minimum value, median, percentage and so on.
2. Measurement data: T-test, rank-sum test and other methods were used.
3, count data: the use of chi-square test, corrected chi-square test, Fisher's exact test and so on.
(V) Statistical analysis report
After the clinical trial is completed, in order to provide a basis for the investigator to write a summary report of the clinical trial, the case report form for the collection of clinical trial data should be sent to a professional data management and statistical analysis organization for statistical analysis of the results of the study. In addition to the statistical analysis of the data from each sub-center (as required by Order No. 5 of the State Food and Drug Administration), the professional data management and statistical analysis organization should combine the data from all centers for statistical analysis and write a summary statistical analysis report. The statistical analysis report includes tables and graphs describing the results of the statistical analysis.
When evaluating the effectiveness of a device, the results of the descriptive statistical analysis should be given for each observation time point (follow-up point). List the test statistics, p-values. For example, the t-test results for the two treatment groups should contain the number of people in each group, mean, standard deviation, median, minimum, maximum, t-value and P-value for comparison of the two groups. For multicenter clinical trials, efficacy evaluations should be adjusted for center and baseline effects (if baseline variables are not balanced between groups).
The safety evaluation of devices is mainly based on descriptive statistical analysis, including the use of devices (duration of use of devices, etc.), the incidence of adverse events and the specific description of adverse events (including the type of adverse events, severity, incidence and duration, and the relationship with the test device, etc.); the changes in the laboratory test values before and after the test, in particular, the changes in the pre-test normal, post-test abnormal and clinically significant conditions; and the changes in laboratory test values before and after the test, especially the pre-test normal, post-test abnormal and clinically significant conditions. Changes in laboratory test values before and after the test, especially normal before the test and abnormal after the test with clinical significance; abnormal changes and their relationship with the test device and follow-up results. If necessary, the significance of differences between groups can be tested.