I. Trial Operation Period [1]
1. Concept: Lack of authorship in RCT will reduce drug effect and statistical guarantee. The simple way to eliminate this effect is to increase the sample size, but this will inevitably increase the test cost. Therefore, researchers can increase the trial period in the research design and avoid this problem by excluding the author's identity before the trial. Probation method refers to understanding the cooperation, compliance and unbearable side effects of subjects through short-term experiments before RCT is randomly divided, so as to exclude subjects who may not be able to adhere to the experiment, and only those who can participate in the experiment are randomly divided in subsequent experiments.
2. The application of trial period method and the results show that: (1) Explore the cooperation of patients: the trial period method can be used to screen patients who may not be able to adhere to the trial before random grouping, exclude patients who cannot adhere to the trial before random grouping, and improve the compliance of the trial.
Physician Health Research is the first large-scale RCT to observe the role of aspirin and β -carotene in preventing coronary heart disease and tumor. After a trial period of 18 months, 33% people who could not persist in the experiment were excluded. During the five-year follow-up observation, it was found that the relative risk of aspirin in preventing coronary heart disease was 0.56(95% confidence limit was 0.45 ~ 0.70). If all the excluded people are included in the trial, assuming that the excluded people are evenly distributed in the two groups and have no therapeutic effect, the recalculated RR value is 0.7 1, which is 25% worse than 0.56. In other words, after the exclusion item is added, it is likely to underestimate or cover up the real curative effect. Therefore, even negative results should be highly valued, and the time limit method has not been tried in RCT. (2) Placebo reaction screening: Placebo reaction refers to the efficacy and side effects of subjects taking placebos in RCT. These phenomena may be the physiological and psychological reactions of patients after taking medicine, and often have a certain impact on RCT results, which can be estimated by trial operation period method. For example, in an RCT for the treatment of depression, 19 people who responded to placebo were excluded after a 7-day trial period. The results showed that there was a significant difference in treatment between the drug group and the placebo group (P=0.04). If 19 people were not excluded, they were randomly divided into two groups, and there was no significant difference (P=0.09). Therefore, excluding those who have placebo response during the trial period can increase the statistical validity of the trial. (3) Screening clinical response: Trial operation can be used to screen the clinical response of patients, such as whether the drug is effective, ineffective, harmful or has side effects. Generally speaking, the selection of RCT patients in the trial period may not fully reflect the effectiveness and safety of the experimental drugs or treatment methods, because the efficacy estimation and statistical test of RCT after excluding the author or those with serious side effects will be different from those in the non-trial period, and there is no effective method to adjust this difference so far.
It is complicated to explain the test effect when RCT is carried out by excluding those who are not authors or have serious side effects during trial operation. If the therapeutic effect and side effects are independent, when the author is excluded due to lack of response to treatment, the therapeutic effect of the trial will be overestimated; When the author is excluded because of poor prognosis, the test results will underestimate the treatment risk; When the author is sensitive to the side effects of the test, the actual side effects rate of the test will be underestimated, and vice versa. If the therapeutic effect is related to side effects, it is more difficult to explain the results. In RCT, sometimes curative effect is related to side effects, that is, people with serious side effects may also have good curative effect. If these objects are excluded, the therapeutic effect cannot be correctly estimated. In fact, in many RCTs, it is difficult to detect the relationship between treatment response and side effects, so it is difficult to estimate the results truly. (4) Promoting clinical application: The trial operation method can also guide clinical application. In 1980s, because many observations showed that there was a high correlation between ventricular ectopic and sudden death, many clinicians began to use antiarrhythmic drugs to inhibit ventricular ectopic in patients after myocardial infarction. Arrhythmia inhibition test (CAST) is an RCT aimed at detecting this clinical practice. After the trial period, eligible patients were randomly divided into drug group and control group, but the CAST trial was terminated soon because of the high mortality rate in the treatment group during the trial period. Due to the use of the trial operation period, the statistical efficiency was improved, and the research results were quickly obtained, which was quickly accepted and popularized in clinic.
In a word, the experimental operation method can improve the compliance of RCT, reduce drug withdrawal and ensure the comparability between the treatment group and the control group, but in the interpretation of the results, it may overestimate the benefits of treatment and underestimate the risks of treatment, and obtain a smaller statistical probability value (P value).
Second, the intention to treat analysis (ITT)
1. Concept [2]: ITT (also called actual test or item effect analysis) was first applied to 196 1, which means that all patients are randomly assigned to any group of RCT, regardless of whether they have completed the test in this group or whether they have really received the treatment in this group, they will stay in the original group for result analysis. The purpose of ITT is to avoid selection bias and maintain comparability between treatment groups.
The simple grouping of RCT is shown in figure 1. In ITT, randomization not only determines the distribution of treatment, but also determines the analysis of patient data.
Figure 1 RCT grouping frame diagram
As can be seen from Figure 1, there will be four groups of patients at the end of the trial. ITT is to compare ①+② group and ③+④ group. Besides ITT, there are other analysis methods. Efficacy analysis (that is, follower analysis, also called explanatory test or biological efficacy test) is to compare Group ② and Group ③, while ignoring Group ① and Group ④. The analysis of receiving treatment is to compare (① transfer group)+③ group and ② group+(④ transfer group) group. These three analysis methods have their own uses, but ITT is the most effective method to evaluate the authenticity of the project. For more information, see the following example.
2. Analysis method example [2]: In a two-year follow-up study of coronary artery bypass surgery, surgical treatment was taken as a new method and drug treatment was taken as a control. Table 1 is the clinical outcome data for two years, and table 2 is the analysis results of the above three methods. ITT analysis showed that the mortality rate of patients in drug treatment group was 7.8%, and that in surgical treatment group was 5.3%, P=0. 17. The therapeutic effects of the two groups were similar. In other analysis methods, the mortality rate of medical treatment group (8.4%) was higher than that of surgical treatment group (4. 1%), P = 0.018; The analysis of receiving treatment was similar to that of the follow-up, and the mortality rate of drug treatment group (9.5%) was higher than that of surgical treatment group (4. 1%), P=0.003.
Table 1 RCT results
Statistical data are allocated to medical treatment and surgical treatment.
perform an operation
Treatment acceptance internal medicine
Receive surgical treatment
Treatment acceptance internal medicine
treat cordially
Number of people who have been followed up for 2 years 48
296
354
20
Death toll 2 27 15 6
Total 50 323 369 26
Table 2 Comparison of three analytical methods
Analysis method assignment group χ2 P
Medical treatment, surgical treatment
Intention to treat analysis 29/373(7.8%)
2 1/395(5.3%)
1.9
0. 17
Dependence analysis 27/323 (8.4%)15/369 (4.1%) 5.60.5438+08.
Treatment analysis: 33/349 (9.5%)17/419 (4.1%) 9.10.003.
It can be seen that the results obtained by the three analysis methods are not consistent. ITT analysis reflects the effects of the two treatment methods after actual clinical application, including various results of patients during the trial; However, when evaluating the real curative effect of the treatment method, ITT analysis will underestimate the therapeutic effect of the test if the test method is really effective. Persistence analysis only analyzes the people who abide by the experiment, and does not completely follow the initial random grouping. In the above example, among the patients assigned to surgical treatment, 26 were converted to medical treatment, and 6 died, with a mortality rate of 23%. These people may have a poor prognosis or die while waiting for surgery; Among the patients assigned to receive medication, the mortality rate of switching to surgery is only 4%. This non-compliance is unbalanced between the two groups. Therefore, when using follow-up analysis, the effect of surgical treatment will be overestimated. Similarly, the therapist overestimated the effect of surgical treatment when analyzing.
3. Application and limitation [3]: The two basic goals of RCT are to obtain the efficacy and effectiveness of the test. The effectiveness of the trial reflects the therapeutic effect in the ideal state, that is, the subjects really accept and complete the treatment. The effect of the trial refers to the actual effect of treatment under general clinical conditions, and the subjects may not comply, change the treatment mode or interrupt the treatment. ITT analyzes and evaluates this result, that is, the actual result of the patient after receiving a certain treatment mode.
As for the effectiveness of the experiment, ITT analysis can get effective information about the effectiveness of the experiment if there are few cases of missing interviews and non-compliance, or if the missing interviews and non-compliance between groups are balanced. However, if it is unbalanced, ITT analysis cannot fully evaluate the efficacy of the trial. If the experimental method is really effective, ITT may underestimate the therapeutic effect, while follower analysis and therapist analysis will overestimate the therapeutic effect.
Therefore, when evaluating the effectiveness of the experiment, ITT analysis, follower analysis and therapist analysis all have certain limitations. A new method, model-based analysis, is proposed, but it cannot completely solve the above problems. In view of this, it is suggested to use the above three analysis methods at the same time to obtain more comprehensive information and make the interpretation of RCT results more reasonable.
Three. Quantitative evaluation index of curative effect
RCT data should be statistically tested first. If the difference is significant, it is still necessary to combine professional knowledge to further judge whether there is a real difference in the effect of the measures. However, only this definite research conclusion is not enough to guide the specific clinical practice. Therefore, it is necessary to choose appropriate indicators to quantitatively express the curative effect. In the past, the relative risk index was widely used. In recent years, absolute risk measurement, especially the number of people needing treatment (NNT) proposed by Laupacis et al. [4] in 1988, has the advantages of being intuitive and easy to understand, and can guide the clinical decision-making of individual patients.
The concept of 1 NNT: NNT refers to the number of patients that clinicians need to treat for a period of time in order to prevent 1 adverse events. It was originally used to evaluate the effect of RCT.
Assuming RCT, patients were randomly divided into treatment group and placebo control group, and followed up for a period of time to observe the occurrence of harmful events in the two groups. Let the incidence of experimental events in the treatment group be EER and that in the control group be CER, then the calculation formula of relative risk index is as follows:
Relative risk, RR), RR = eer/cer formula 1.
Validity index, IE), IE = CER/ energy efficiency ratio formula 2.
Relative risk reduction (RRR) is also called protection rate, RRR=(CER-EER)/CER Formula 3.
Absolute risk reduction (ARR), arr = cer-eer formula 4.
Mathematically speaking, the number of people needing treatment NNT is equal to the reciprocal of the absolute risk reduction value, that is, NNT = 1/ARR formula 5.
2. Comparison of NNT with other indicators: Cook and others [5] take a paper on antihypertensive treatment of mild to moderate hypertension as an example to illustrate the advantages of NNT compared with other indicators. In this study, patients were divided into mild hypertension (diastolic pressure ≤ 110mmhg (1mmhg = 0.133kpa)) and moderate hypertension (diastolic pressure ≤115mmhg). Patients in each layer were randomly divided into two groups: antihypertensive drug group and placebo group. Taking the occurrence of stroke as the observation end point. After 5 years of follow-up, it was found that the incidence of stroke in the control group and the antihypertensive treatment group was 20% and 65438 02% respectively. The incidence of mild hypertension in the two groups was 65438 0.5% and 0.9% respectively (Table 3).
Table 3 Analysis of curative effect of antihypertensive therapy in patients with hypertension
hypertension
Incidence rate of classified stroke (%) rrrrnnt
control group
(CER) treatment group
(EER)
Medium 0.20
0. 12
0.60
0.40
0.08
13
Mild 0.0150.009 0.60 0.40 0.006167
In this study, the stroke risk (also known as baseline risk) of untreated patients with moderate hypertension is 13 times that of mild patients, but the RR of both patients is 0.60, and the RRR is 0.40. It can be seen that the relative risk index does not consider the patient's past medical history, nor can it reflect the risk of untreated. In clinical practice, it is very important to consider these factors before making a treatment decision. For example, for patients with moderate and severe hypertension, taking some antihypertensive drugs can reduce the incidence of stroke by 40%, that is, the protection rate is 40%, which will be statistically significant and clinically important. But for patients with mild hypertension, reducing the risk by 40% may not be enough to offset the side effects and treatment costs. Therefore, when the baseline risk of adverse events is low or high, only using the relative risk index will overestimate or underestimate the absolute impact of treatment.
The absolute risk index takes into account the difference of patients' baseline risk. For example, the ARR of patients with mild hypertension in this case is 0.08 and 0.006 respectively, which is 13 times compared with them. However, the index is expressed in decimal or fractional form, which is difficult to be understood by doctors and patients and difficult to be used in clinic. The reciprocal of ARR, i.e. NNT, is about 13, which shows that in order to prevent 1 case of stroke, doctors need to treat 13 cases of moderate hypertension for five years, which is more intuitive and acceptable than ARR=0.08. In addition, we can also see the superiority of NNT relative to the relative effect evaluation index from the comparison of patients with mild to moderate hypertension. The protective rate of antihypertensive therapy for both types of patients is 40%, which seems to indicate that the two groups of patients should be given the same intensity of treatment. However, in order to prevent 1 stroke, only 13 patients with moderate hypertension need treatment, while 167 patients with mild hypertension need treatment. Obviously, this will lead to different treatment decisions.
NNT = 1/(RRR× CER) can also be deduced from formulas 3, 4 and 5. It can be seen that both baseline risk and relative risk reduction have an impact on NNT. If the measures with low protection rate are applied to people with high accident rate, NNT can be lower and more benefits can be obtained. For example, in diseases with a baseline risk of 60%, only 10% RRR can get 17 people who need treatment. Conversely, even if the protection rate of a measure is high, if it is used for people with low incidence rate, the benefit is limited. For example, if the baseline risk is 10%, the RRR needs to be 60% to get 17 people who need treatment.
3. Calculation of NNT confidence interval: As a point estimate, NNT also has a 95% confidence interval, and its calculation is very simple, which is equal to the reciprocal of ARR's 95% confidence interval. For example, the ARR and 95% confidence interval of a drug are 10% (5% ~ 15%), then NNT = 10, and the 95% confidence interval is 6.7 ~ 20. However, when the treatment is ineffective, for example, the ARR is still 10%, and the 95% confidence interval is wide and includes 0, ranging from-5% to 25%. On this basis, NNT = 10 (-20 ~ 4) is calculated. There are two problems with the confidence interval calculated by this method. First, the lower limit is negative; Second, it does not include the optimal point value of 10. In order to avoid this contradictory result, some scholars suggest that it is not necessary to calculate the confidence interval of NNT when there is no significant difference between the two groups of treatment measures. Altman [6] proposed a solution to the above contradiction in 1998. When NNT is negative, it means that the treatment measures have harmful effects. Therefore, NNT can be decomposed into NNT(H) and NNT(B). NNT(H) refers to the number of people who need treatment, which produces 1 more harmful effects (hazards) than no treatment, and can be used to indicate the size of side effects. NNT(B) refers to the number of people who need treatment more than those who don't receive treatment 1 case. In addition, when the measures are ineffective, arr = 0, then NNT is infinite, and the confidence interval of NNT should also include infinity. In this way, the above confidence interval can be described as (NNTB4~∞~NNTH20).
4. Extended application of NNT: In addition to long-term drug trials, NNT can also be extended to evaluate other clinical methods. For example, the number of people who need surgical prevention 1 harmful events; The number of people who need to be vaccinated against 1 infection; The number of people who need early screening and diagnosis to prevent 1 cancer death in n years; If 1 harmful events occur, the number of people who need to be exposed to certain risk factors, etc. In addition, based on NNT, we can also calculate the direct cost of preventing 1 adverse events, that is, cost minimization analysis, so as to evaluate the effects of various preventive measures in health economics, and thus better guide clinical decision-making and the choice of the best intervention strategy for public health projects.