This is the definition of Immortal time bias on the Catalog of Bias
“Immortal time bias: A distortion that modifies an association between an exposure and an outcome, caused when a cohort study is designed so that follow-up includes a period of time where participants in the exposed group cannot experience the outcome and are essentially ‘immortal’.”
A retrospective cohort study aimed to estimate all-cause-mortality of patients with a specific disease. one of the inclusion criteria to recruit patient is to have a minimum of 12 months followup period after the diagnosis of the disease. Is this “immortal time bias” or “selection bias”?
Looking at the definition mentioned above and substituting with input from the study design in Question…
“A distortion that modifies an association between an exposure (Disease) and an outcome (Mortality), caused when a cohort study is designed so that follow-up includes a period of time where participants in the exposed group cannot experience the outcome (minimum followup of 12 months) and are essentially ‘immortal’ (if the event happen before 12 months the patient will be excluded).”
Kindly advise if the mentioned study design will introduce immortal time bias.
What you are referring to is qualification time in a landmark analysis. The way qualification time works is that it provides a requirement that all treatment groups must satisfy, and should be applied in an equal-opportunity fashion. Then the clock starts over and prospective follow-up for outcomes starts.
It’s far from foolproof. Defining time zero and qualification times in observational research, besides observational studies disrespecting the need to blind patients and providers to treatments assigned, are one of the greatest problems in observational research.
What you described on its own is indeed not immortal time bias, which is one of the most pernicious and impactful biases in survival analyses. Not accounting for immortal time bias can often give you the opposite results of what is truly happening in the dataset.
For immortal time bias to happen, one or more of your comparator groups would have to be defined by an event that occurs after the timepoint you have set as time zero. For example, in this study we retrospectively assessed the impact of performing cytoreductive nephrectomy (CN) in patients with kidney cancer treated with immunotherapy. The immortal time bias can thus occur because patients in our dataset often underwent CN many months after starting immunotherapy. Therefore, a standard analysis would have been biased in favor of the CN group. We therefore used a time-dependent Cox regression (as described here) to account for this immortal time bias.
Here is also a nice article describing immortal time bias using causal diagrams. And another good one here.