Inspired by Shih C-C et al QJM 2019.
This paper struck me as interesting because of its size, its overlap with previous topics on the blog (VCIND, CHD in RA), but most of all because it introduced me to the marvellous phenomenon called ‘immortal time bias’. I was so intrigued that I was even motivated to read two papers from hardcore epidemiology on the subject.[2,3] The reality of immortal time bias is nowhere near as appealing as the name, but it does just manage to maintain a passing (:D) interest to non-epidemiologists. I’ll try to explain later on, but first let’s have a look at the paper itself.
It is a further observational study coming from Taiwan, where pretty much the entire population is on a national health system database. Shih et al decided to look at post stroke cognitive impairment (PSCI) – there has to be an overlap somewhere here with VCIND, but let’s not confuse ourselves with acronyms for now. Out of a population of 23 million between 2000 and 2007, they found 67521 newly diagnosed cases of PSCI. Of this group, 6661 had been treated with TCM (acupuncture and or herbs). This group was then matched (by propensity score for the outcome of interest) with an equal number who had not received TCM treatments.
The outcome of interest was the requirement for emergency care or hospitalisation. This was reduced in the group who received TCM treatments but note that this is an observation of association and not evidence of causation.
Edited from the results in the published paper:
The adjusted rate ratios (RRs) of emergency care, hospitalization and ICU care associated with TCM were 0.87 (95% CI=0.82-0.92), 0.81 (95% CI=0.78-0.84), and 0.67 (95% CI=0.61-0.73) respectively. The PSCI patients treated with a combination of acupuncture and herbal medicine generally had the lowest risk. RRs were better for herbal medicine across all outcomes, but it is worth noting that the lowest RR reported was 0.53 (95% CI=0.45-0.62), and this was for ICU care in patients having both acupuncture and herbal medicine.
ICU stands for intensive care unit of course, and you don’t often see that in an article with immortal in the title!
Herbal medicine was more popular, with 4180 patients using only herbal medicine, 755 using only acupuncture, and 1726 using both.
So what about immortal time and its biasing influence! Note that I am refraining from giving it an acronym for fear of confusing the physical medicine types out there ;-).
If you are a patient included in a study like this and you are in the group of interest ie a patient who uses TCM treatment, the time from enrolment (diagnosis of PSCI) to your first use of TCM is called immortal time, and unless it is controlled for, it can bias the results. Why is it immortal time? Well, by selection of a patient who has PSCI and then goes on to receive TCM that patient must have survived from diagnosis to first treatment, so during that time they were technically (for the purposes of the study) immortal. If they had died they would not have entered the study. But the immortal time here belongs to the control group not the treatment group since treatment has not started. So there you have it. Epidemiologist have to control for this immortal time so that the results are not biased in favour of the factor used to select the group of interest.
It’s a great term, but now it has lost its mystery for me I’m afraid, and for any of you that have read this far ;-). Well that is an inevitable part of learning I suppose. Just think back to what was going through your mind when you read the title – immortal time…
1 Shih C-C, Yeh C-C, Yang J-L, et al. Reduced use of emergency care and hospitalization in patients with post-stroke cognitive impairment treated with traditional Chinese medicine. QJM An Int J Med Published Online First: 18 February 2019. doi:10.1093/qjmed/hcz044
2 Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 2008;167:492–9. doi:10.1093/aje/kwm324
3 Lévesque LE, Hanley JA, Kezouh A, et al. Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes. BMJ 2010;340:b5087. doi:10.1136/bmj.b5087