I am interested in using mean cumulative functions for comparing adverse event burdens between randomized groups in drug trials. Specifically, to better specify cumulative adverse event burden with time, adverse events can be weighted according to their severity. I use the CTCAE severity scale (mild = 1, moderate = 2, severe = 3, life-threatening = 4, and death = 5) to grade adverse events, which seems straigtforward and clear. However, the statistical weighting of these grades should logically be related to 1-minus-the- probability-of-survival-with-a-good-quality-of-life. I'm expecting a weighting system that looks like the following based on my own, arbitrary, best guess: Grade 1 Mild: 0.01; Grade 2 Moderate: 0.1; Grade 3 Severe: 0.60; Grade 4 Life Threatening: 0.9; Grade 5 Death: 1. I have not been able to find scientific studies addressing this issue, or providing a non-arbitrary way to weight Adverse Events. Does anyone have knowledge on this subject?
I am somewhat unsure if this type of analysis will give meaningful results in that you are loosing a lot of information by quantifying the importance of each AE like this. Also while death certainly means no QOL, the range of likely QOL decrease for each of the other classes will be wide. Outcome information (recovered or recovered with sequelae) should probably be weighted in as well since a moderate AE with a permanent loss of function will affect the long.term QOL more than a life threatening event causing hospitalisation but complete recovery.
If you work with data in MedDRA terms, there is the Important Medical Events (IME) and Designated Medical Events (DME) lists that always gets counted as significant events when evaluating possible adverse drug reactions signal (see EMA guidance on signal evaluation). These terms might be used in an analysis as additional information in addition to severity.
See Link below
For each variable you find criteria how to assign grades.
http://evs.nci.nih.gov/ftp1/CTCAE/CTCAE_4.03_2010-06-14_QuickReference_5x7.pdf
Thanks for the responses! To summarize, elements that should be incorporated into hypothetical weights for Grades 1 to 5 are:
1, impact on quality of life (hard to get! unless....)
2, recovered versus recovered with sequelae (a modulator for the weight maybe?)
3. certain MedDRA classifications.
I'm guessing I should also go look at the type of "action taken" in response to each AE.
Thanks again for the help, and throw in more ideas if you have them. -Carey
Just a thought: Are AE reported data from RCTs valid to start with? An RCT with precise AE estimates would require large sample sizes. E.g when you calculate AE RiskRatios.
Hi Frederik, in response to your question: Certain fragile populations can have large numbers of AEs. I'm looking at various situations where mortality is high (over 50%) for the time frame considered, and survivors can have considerable and variable AE burden. Keep in mind that I am looking at all AEs, not just unexpected ones and not just those imputed to a drug. In a more generalized fashion, I feel that whatever the situation, given the plethora of data included in a basic AE description, I find it quite unbearable to just tabulate them. The vast majority of information therein is also lost when simple grouped frequencies are just counted. As a previous post mentioned, not all AEs are equal, and they represent variation that I think deserves more attention. -Carey
Hello, I think that when you meant grades 1 to 5 you mentioned NCI-CTC AE tables (available in the internet), which grades all AEs from 1 (mild )to 5 (death) degrees. I suppose that you should probably use waterfall plot and separately evaluate AEs of each severety between groups . At least this approach shows what has changed after administration of study treatment. You should also evaluate what % of study subjects did not develop any AEs, any SAEs. The point is that rather frequently the same patient may develop multiple AEs/SAEs throughout the study. in this case also interesting compare the % of patients who did not develop the AEs between groups.
You should also evaluate patterns of AEs/SAEs in each group using standard SOC assignment (system-organ-class). In all cases there multiple ways to toy with the data. Good luck!
I think that the original question is the wrong question. Rather than asking "How do I do this?", you might ask "Is it appropriate to do this?" I believe that it is not. For example, in the National Cancer Institute's CTCAE, anaemia is labelled Grade 2 if the haemoglobin is 8-10 g/dl and Grade 3 if the haemoglobin is 6.5-8 g/dl; death is Grade 5. So, if you give a Grade 2 reaction a weighting of, say, 0.25, a grade 3 reaction a weighting of 0.5, and a grade 5 reaction a weighting of 1, you are implying that four cases of mild anaemia and two cases of moderate anaemia are as bad as one death. Furthermore, the probability of survival with a good quality of life is very high and much the same for Grades 1-4 in this case, which may or may not be so in other cases. The allocation of numbered grades to these categorical descriptions is misleading. If they had been labelled A, B, C, D, and E the question of possible weighting would probably not have arisen.
I find myself in agreement Svendsen and Aronson. There are too many factors baked in to a trial design, and likewise too much heterogeneity in patients, responses, etc, for this to be a meaningful analysis. In my experience in clinical trials and as a physician, AEs reported in that setting tend to be over-reported and clarity is lacking except for spectacularly terrible AEs until after marketing (if the product is cleared). The AE literature and debates around anti-TNF targeting and AEs is very informative in this regard.
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Medical Pharmacology
- Clinical Pharmacology