One of the purposes of various clinical studies before and after new drugs are launched is drug safety evaluation. The indicators for safety evaluation include measurement of vital signs, laboratory tests, and observation of adverse events. The current analysis of adverse events is limited to describing the incidence of adverse events and enumerating a list of adverse events, which cannot yet meet the needs of safety evaluation in clinical research. This article attempts to introduce a visual method into the analysis of adverse events from the perspective of statistical description. This method It converts the observed data of adverse events into static or dynamic images and graphics, discovers drug safety signals hidden in the data, and provides help for understanding the data, analyzing the data, forming concepts, and finding patterns. This article displays a total of 6 types of graphics: forest diagram, tree diagram, volcano diagram, Venn diagram, survival function diagram and risk function diagram. We will discuss the graphic design and interpretation based on the purpose of adverse event analysis.
1. Research Background
1.1 Adverse event definition
An adverse event (AE) is defined as an adverse medical event that occurs after a patient or clinical trial subject receives a drug, but does not necessarily have a causal relationship with the drug. Among them, specific adverse events that occur during clinical trials such as requiring hospitalization, prolonging hospitalization, causing disability, affecting work ability, endangering life or death, and causing congenital malformations are defined as serious adverse events (SAE). During the normal use of drugs at prescribed doses, adverse events that are harmful but not expected and have a causal relationship with the use of the drug are called adverse drug reactions (adverse reactions, ADR).
1.2 Recording of adverse event information
There are established standards for recording adverse events in clinical trials, and it is currently recommended to use the Clinical Data Exchange Standards Association ( Clinical Data Interchange Standards Consortium , CDISC) standards proposed. Clinical data acquisition standards in CDISC ( Clinical Data Acquisition Standards Harmonization , CDASH) pointed out that the collection of adverse events must include the following content: name of the adverse event, start time, end time, whether the adverse event is serious, relationship with the drug, measures taken, outcome and severity. Information such as adverse event classification, adverse event patterns, adverse event coding, and adverse event system classification can be selectively collected as needed.
1.3 Adverse event analysis
The current simplest method of analyzing adverse events is based on the crude rate obtained by dividing the number of patients who experienced an adverse event by the total number of patients who received study treatment. The rough rate can reflect the basic situation of the occurrence of adverse events, but it is difficult to reflect the details and characteristics of the adverse events. Another shortcoming of the rough rate calculation is that the exposure time is not taken into account. In order to take the exposure time into account, we can examine the events per unit of exposure time. Occurrences are defined as the ratio of the number of occurrences of a specific event to the total time of each patient's exposure from the time of occurrence or the end of the trial. The conventional practice of analyzing adverse events in general or individually cannot easily obtain a complete picture of safety. Some scholars have proposed classifying safety data according to body systems. The format for listing adverse events given in ICH-E3 requires not only describing the total adverse event experience, but also listing all adverse events. Each adverse event should be listed in the statistical table, and the corresponding number and rate of each group should be listed; adverse events should also be classified according to body systems; whether the adverse events are related to the study treatment should be distinguished; if necessary, adverse events should also be classified according to the severity of the adverse events. . Adverse events with higher incidence and greater differences between groups should be described in detail.
1.4 Visual evaluation of adverse events
Systematic and comprehensive adverse event analysis is carried out through some visual methods. Graphics can concisely convey various aspects of information, especially based on some new technologies developed in recent years, such as forest diagrams, tree diagrams, volcano diagrams, Webster diagrams, etc. Graphics convey information more effectively than tables. Furthermore, graphics can integrate statistical descriptions of data with their corresponding statistics. In exploratory clinical trials, graphics can illustrate the discovery of abnormal safety signals, and in confirmatory clinical trials, graphics can illustrate our interpretation and elaboration of known safety issues.
Visual representation of adverse events has many advantages. First, it reduces the likelihood of missing signals and increases the likelihood of early detection. Second, from a human physiological point of view, graphics stimulate the various potentials of the human eye—fast, high-frequency recognition and processing, recognition and association of relevant content, and achieve the goal of processing information temporarily and effortlessly. Purpose, when a graphic is presented, it can directly enter people's psychological space without much explanation and reasoning process. Third, well-designed graphics improve the statistician's ability to communicate with other colleagues on the clinical research project team, including physicians. Fourth, statistical graphics can not only convey information faster and more effectively, but also make the information conveyed by graphics more acceptable to non-statisticians who are not involved in the design and expression of tables of statistical results. Visualizing statistical results can improve the quality of collaboration during clinical trials and shorten time and decision-making, especially when the research problems are complex. Some authors have written specifically on graphic design.
The visual evaluation of adverse events should have the following key elements: 1. Integrate the necessary statistics into the graph, 2. Add appropriate indicator lines to the graph, 3. The graph can visually compare the differences between the test group and the control group, 4. Facilitate comparison between adverse events, 5. The graph should highlight the adverse events that require focus.
- Data sources and drawing software:
The data of this study come from a randomized double-blind placebo-controlled phase III clinical trial for the treatment of patients with subarachnoid hemorrhage. The mechanism of action of the drug evaluated is mainly to inhibit smooth muscle contraction and dilate blood vessels by blocking calcium channels. A total of 906 patients were enrolled in the study, including 449 patients in the drug group and 457 patients in the placebo group. Factors that may affect the prognosis were balanced between the two groups. Patients in both groups continued to take the drug for 14 days, 0.15 mg/Kg/hr per day, and were followed up. 120 days of observation results.
JMP Clinical is a statistical software developed specifically for clinical trial data analysis. It uses standard CDISC data format, combines advanced statistical methods with rich interactive graphics, and provides powerful statistical analysis capabilities in a user-friendly environment. . SAS itself also has drawing capabilities. The graphics drawn in this article are all using SAS and JMP Clinical.
3. Visual representation of adverse events
3.1 Describe the incidence of adverse events and relative risks between groups ( Relative Risk, RR )
The incidence rate of adverse events is the most commonly used measure to describe the occurrence of adverse events, and RR reflects the difference in the incidence of adverse events between the trial combination control groups. Using the principle of forest plot, the incidence rate of adverse events, RR and its 95% confidence interval are integrated into one figure. The incidence rate of adverse events in each group can be observed, and the group can be inferred based on the RR and its 95% confidence interval. Whether the differences between the two are statistically significant, and comparisons between different types of adverse events can also be conducted longitudinally. According to the research purpose, adverse events are sorted according to certain rules in the forest diagram, which can intuitively and prominently display the adverse events that require the most attention. For example, in clinical research, the purpose is to describe adverse events with a high incidence, and they should be sorted from high to low incidence.
A total of 201 adverse events were observed during the study , This article selects adverse events with an incidence rate of ≥5% in the trial group and draws Figure 1 in order of incidence from high to low. The scatter points on the left panel of the figure represent the frequency of adverse events, the red triangle represents the test group, and the blue dots represent the control group; the right panel represents the RR of the adverse event test group relative to the control group and its 95% confidence interval. The scale on the X-axis is log 2 Conversion, the position of the dotted line is scale 0, which is equivalent to RR=2 0 =1; the position of scale 1 is equivalent to RR=2 1 =2, and so on. It can be observed from the figure that the adverse events with higher incidence rates in the experimental group than in the control group and with statistically significant differences are "phlebitis", "isotonic urine", "hypotension" and "pulmonary edema"; in the control group, The adverse events with a higher rate than the experimental group and statistically significant differences were "hypertension", "vasoconstriction" and "hyponatremia".
Drawing according to the principle of forest plot, each adverse event needs to be treated as a column, and there may be hundreds of adverse events in a clinical trial. Therefore, in practical applications, some adverse events should be selected for visualization purposes. Drawing, selection criteria should be carefully considered. In this article, common adverse events with an incidence rate of ≥5% are selected. There is another problem in this selection: there may be some adverse events with an incidence rate of <5%, but the difference between the groups is statistically significant. If based on the incidence rate, Screening may miss such adverse events.
Figure 1 Adverse event incidence and relative risk graph
3.2 Describe body system classification, number of adverse events, and differences between groups
Systematic classification of adverse events is a necessary part of adverse event analysis. Currently, the incidence of adverse events of each system is often described in tables. Tree diagrams can be used to systematically classify adverse events, the incidence of each adverse event, and the statistics of differences between groups. Meaning integrated into a picture. Tree diagram is a representative technology based on hierarchy visualization technology. Its basic idea is to divide the n-dimensional data space into several subspaces, and these subspaces are still organized in a hierarchical structure and represented graphically.
Figure 2 is drawn based on the above principles. Each square in the figure represents an adverse event. Adverse events for the same system are grouped together and semi-transparently labeled with their system classification name. The size of the box is proportional to the number of such adverse events. The larger the box, the greater the total number of such adverse events. On the contrary, the smaller the box, the smaller the number of such adverse events. The color of the square is red, which means that the number of adverse events in the test group is greater than that in the control group, and green means that the number of adverse events in the control group is greater than that in the test group. The color of the box and the number of adverse events are determined by the P value of the difference between the groups. The smaller the P value and the darker the color, the higher the significance of the difference between the groups.
Figure 2 Tree diagram
This graphical expression can give readers a strong visual impact. Readers can directly see the darkest square. The difference between the adverse event groups is very significant. This difference is likely to be caused by the treatment measures. In addition, readers can also Intuitively observe information such as the total number of adverse events and differences between groups of adverse events. Obtaining this information is simple and direct, which is incomparable with digital tables. It can be observed from the figure that the incidence rates of "nervous system diseases" and "vascular diseases" are roughly the same, but the significance of the difference between the groups of "vascular diseases" is much higher than that of "nervous system diseases". "Vascular disease" mainly refers to four adverse events, among which "phlebitis" and "hypotension" have a higher incidence rate in the experimental group than in the control group; while "vasoconstriction" and "hypertension" have a higher incidence rate in the control group higher than the experimental group.
Figure 2 is drawn by JMP. In this software, you can directly click on the adverse event box of interest to perform statistical description or statistical inference on the adverse event to obtain the specific values required. The purposes of clinical research are often different. In practical applications, appropriate transformations can be made on the basis of Figure 2, for example, two tree diagrams can be output in groups, etc.
3.3 Screening for adverse events with statistically significant differences between groups
Screening adverse events with statistically significant differences between groups and clarifying their relationship with medication is one of the main purposes of adverse event analysis. The application of volcano plots can help researchers quickly find adverse events with statistically significant differences between groups, and then Conduct a detailed analysis of these adverse events. In statistics, volcano plots are used to quickly identify differences between groups of the same type of data in large data sets in the form of scatter plots. Drawing Figure 3 based on the above principles, it not only highlights the adverse events with significant differences between groups, but also displays the overall incidence rate of these adverse events and the risk difference (RD) between groups.
The X-axis in the figure represents the difference in the rate of adverse events between the experimental group and the control group (experimental group - control group), and the Y-axis represents the statistical significance of the difference between the groups. Points with an RD greater than 0 indicate that the adverse event occurred more frequently in the test group, and points less than 0 indicate that the adverse event occurred more frequently in the control group. The farther away from the 0 point, the greater the difference between the groups. The Y-axis represents the significance of the difference between groups. The larger the value of the point on the Y-axis, the smaller the P value of the difference between the adverse events between groups. The statistically significant difference is calculated through the SAS mixed model and uses the statistical method False Discover Rate (FDR) corrected for multiple testing. After adjustment, adverse events above the red dotted line have statistical differences between groups. The meaning of the color of the dots is as shown in the legend, and the size of the dots represents the incidence of adverse events.
Figure 3 Volcano plot reflecting differences in adverse events between groups and the size of the differences
3.4 The combined occurrence of adverse reactions
Through the volcano chart, it was observed that there are five main types of adverse events that need to be focused on. At this time, researchers often need to analyze these five adverse events in detail, including inferring the causes of their occurrence by observing the concurrency of adverse events. For this purpose, a Venn diagram or Venn diagram can be used. In statistics, a Venn diagram is a graph that uses a closed curve to intuitively represent a set and its relationships.
Figure 4 Venn diagram
3.5 Long-term follow-up clinical studies require AE survival analysis
In the statistical analysis of adverse events, in order to observe the risk of a subject's occurrence of an adverse event over time, the adverse event data can be used as time-to-event data for survival analysis. It is very appropriate to use the survival analysis model to evaluate the risk of events, because the adverse event data itself includes both the occurrence (with or without) and the time of event occurrence. In long-term treatment, there will be a considerable proportion of censored values, so Survival analysis should be used to calculate the cumulative rate of adverse events to avoid underestimation of risk. ICH recommends the use of survival analysis methods to analyze adverse event data. Foreign literature has demonstrated the significance of survival analysis models in adverse event analysis, and discussed the design of survival curves. There are many types of adverse events in clinical studies, and this method can only perform survival analysis on a certain adverse event that focuses on it.
The analysis of adverse events focuses on the relationship between the time of adverse events and the time of medication. Figure 5 is the Kaplan-Meier curve of the occurrence time of phlebitis in the two groups. The numbers at the bottom of the figure represent the number of people at the beginning of each visit. The " +” indicates that there are censored cases at this time, and 95% confidence intervals are drawn above and below the curve. If the two curves are very close, and then draw the confidence interval, it will be difficult to distinguish the two curves. In order to keep the graph simple and easy to distinguish, it is recommended to adopt the following methods: 1. Use the standard error instead of the 95% confidence interval, or just Draw the standard error above the curve. 2. Move the experimental group curve to the left and the placebo group curve to the right so that the credible intervals of the two curves are staggered. 3. No longer draw the credible intervals. Mark HR and Its confidence interval or the P value of the Log-rank test can be used.
Figure 5 Kaplan-meier curve of time to occurrence of phlebitis
It can be seen from the figure that phlebitis occurred within 17 days after the start of the study, mainly within the first 14 days, which is consistent with the medication time of the study, and the occurrence rate of phlebitis in the test group was much faster than that in the placebo group, suggesting that This drug may induce phlebitis during use.
In addition to the survival function, the risk function can also be calculated through the life table method in survival analysis. Figure 6 plots the risk function of phlebitis during follow-up. Likewise, the beginning headcount is marked below the graph, and the standard error is plotted on the curve. The graph is calculated using the life table method, and two days are set as an interval. The risk rate of each interval is the risk rate at the midpoint of the interval. This graph directly reflects the risk of phlebitis in subjects at different times after taking the drug. It should be noted that since the life table method in SAS assumes that all censored cases occur at the midpoint of the time interval, which is equivalent to the censored cases only experiencing half of the time interval, so when calculating the number of people at the beginning of the period, the censored cases Multiply by 0.5. This will result in the number of people at the beginning of the period not being an integer.
Figure 6 Adverse event risk function diagram
In the evaluation of adverse events in clinical trials with long-term follow-up, it is recommended to use the above two figures.
4. discuss
There are many directions for analyzing adverse events in clinical trials. This article tries to analyze and display it from multiple angles, but it still cannot cover everything. The above 6 pictures can not only be used for the analysis purposes shown in the article. For example, in clinical trials, we will encounter the question "If the patient has been treated for X months without adverse events, can we relax our vigilance? How long must we maintain it?" vigilance?" According to this need, Figure 5 can be used to make a risk function diagram of the first occurrence time of adverse events, and then it can be observed from the figure that the risk rate drops to 0 after X months, that is, within X months Be alert. In short, in the process of visual analysis of adverse events, the purpose of visualization must be combined with the characteristics of the graphics to select the most appropriate graphics.
It is common for the frequency of adverse events in one of the two groups of clinical trials to be 0. For this type of data, a common method is to add 0.5 to each grid and then calculate the RR, or risk difference (RD). In this article, rate difference is used for statistics.
In the analysis of adverse events, some potential influencing factors of adverse events should also be considered, such as course of disease, age, gender, severity of illness, dosage of medication, etc. Logistic regression analysis can be used to use relevant factors as covariates to calculate the OR value. Use OR Values plotted in Figure 1. The three characteristics of adverse events: 0 frequency has a large proportion, the variance of the frequency of adverse events is greater than the mean, and the clustering of adverse events occurs, that is, the probability of another adverse event changes after an adverse event occurs, make it more suitable to use negative events. Binomial regression was used to fit the model. Therefore, when logistic regression is not applicable, the negative binomial regression model can be considered.
If statistical testing is performed on inter-group differences in adverse events in clinical trials, the problem of multiple comparisons must be taken into consideration to reduce false positive errors. Traditional multiple comparisons are based on controlling the total type I error rate, but the results are too conservative when there are too many groups. The method based on controlling the error rate of positive results can not only obtain higher test performance, but also better control the total type I error rate. Error rate.
5. Summarize
Before a new drug is put on the market, its effectiveness and safety must be confirmed through scientific, standardized clinical trials in accordance with GCP requirements. The analysis of adverse events is an important aspect of safety evaluation, so it is particularly important. However, the current analysis of adverse events is not perfect. Of course, this is also related to the characteristics of safety analysis, because safety evaluation is different from effectiveness evaluation. For effectiveness evaluation, clinical trial endpoints must be clearly defined in the research plan in advance. Hypothesis testing compares results between different groups to focus on the average response of the target population; safety endpoints are usually not determined in advance, and safety analysis tends to be exploratory in nature. Compared with clinical efficacy analysis, the average value is valued; safety evaluation is more focused on individual responses. This determines that adverse event analysis is mainly conducted in the form of statistical description.
However, the current statistical description is mainly based on tables. The large number of tables makes it difficult to effectively convey information. The visual method is used to analyze adverse events, which is not only concise and has clear target instructions, but also integrates relevant main statistics into graphics to improve communication efficiency and save research and development time. Therefore, in the analysis of adverse events in clinical trials, it is recommended to use visual analysis methods to discover potential safety signals by drawing different statistical graphics.
