On 2015 Mar 05, Anders Prior commented:

High effect sizes and immortal person-time

In a paper published in Medicine, Wang et al. presented this cohort study of 47,225 stroke patients.¹ (Wang JY, 2014) They concluded that the 30-day mortality after stroke was significantly reduced if treatment with antipsychotic drugs was initiated before or after the stroke.

The reported effect sizes were very large; the mortality rate in stroke patients decreased by 73% in those who received antipsychotic treatment before the stroke and by 87% in those who received antipsychotic treatment shortly after the stroke. Effects of this size are rarely seen in such studies.

The study was based on a nested case-control design. Included cases were stroke patients who died within 30 days after having a stroke, while controls (matched on age, gender and stroke date) were stroke patients who survived at least 30 days after their stroke. The main exposure was antipsychotic drugs given at any time within 30 days after the stroke date. Multivariate logistic regression was used to calculate odds ratios of mortality.

We are concerned that the results may be substantially biased because the definition of the antipsychotic user group conditions on the future; in order to receive the antipsychotic treatment and become a member of the antipsychotic user group, study participants need to survive until the drug is prescribed. In other words, study participants in the antipsychotic user group are ‘immortal’ until the day of treatment (immortal person-time).² The authors do not describe any analytical measures taken to counteract this conditioning. If they have not taken this into account, this would pose a very serious problem as also described numerous times in the epidemiological literature.³ (Hanley JA, 2014) Immortal time bias will generate an illusion of treatment effectiveness and is frequently found in observational studies that compare with non-users.⁴ (Suissa S, 2007)

In general, suspicion for immortal time bias should be raised when exposure groups are assigned with no regard to exposure time in a longitudinal study. Furthermore, the reported effect sizes are surprisingly high, especially when considering the fragility of the population in question. This group consists of stroke patients with complications; they may suffer from e.g. post-stroke delirium and may need antipsychotic treatment. They would most likely have more adverse outcomes, not the opposite.^5,6 (Shi Q, 2012, Prior A, 2014)

Dr. Anders Prior, MD

Research Unit for General Practice and Section for General Medical Practice, Department of Public Health, Aarhus University, Denmark

Dr. Thomas Munk Laursen, PhD

National Centre for Register-based Research, Department of Economics and Business, Aarhus University, Denmark

Prof. Mogens Vestergaard, PhD

Research Unit for General Practice and Section for General Medical Practice, Department of Public Health, Aarhus University, Denmark

References

1 Wang JY, Wang CY, Tan CH, Chao TT, Huang YS, Lee CC. Effect of different antipsychotic drugs on short-term mortality in stroke patients. Medicine (Baltimore). 2014;93(25):e170.

2 Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 3rd ed. Philadelphia, PA: Lippincott Williams & Wilkins; 2008.

3 Hanley JA, Foster BJ. Avoiding blunders involving 'immortal time'. Int J Epidemiol. 2014;43(3):949-961.

4 Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf. 2007;16(3):241-249.

5 Shi Q, Presutti R, Selchen D, Saposnik G. Delirium in acute stroke: A systematic review and meta-analysis. Stroke. 2012;43(3):645-649.

6 Prior A, Laursen TM, Larsen KK, et al. Post-stroke mortality, stroke severity, and preadmission antipsychotic medicine use--a population-based cohort study. PLoS One. 2014;9(1):e84103.

On 2015 Mar 05, Anders Prior commented:

High effect sizes and immortal person-time

In a paper published in Medicine, Wang et al. presented this cohort study of 47,225 stroke patients.¹ (Wang JY, 2014) They concluded that the 30-day mortality after stroke was significantly reduced if treatment with antipsychotic drugs was initiated before or after the stroke.

The reported effect sizes were very large; the mortality rate in stroke patients decreased by 73% in those who received antipsychotic treatment before the stroke and by 87% in those who received antipsychotic treatment shortly after the stroke. Effects of this size are rarely seen in such studies.

The study was based on a nested case-control design. Included cases were stroke patients who died within 30 days after having a stroke, while controls (matched on age, gender and stroke date) were stroke patients who survived at least 30 days after their stroke. The main exposure was antipsychotic drugs given at any time within 30 days after the stroke date. Multivariate logistic regression was used to calculate odds ratios of mortality.

We are concerned that the results may be substantially biased because the definition of the antipsychotic user group conditions on the future; in order to receive the antipsychotic treatment and become a member of the antipsychotic user group, study participants need to survive until the drug is prescribed. In other words, study participants in the antipsychotic user group are ‘immortal’ until the day of treatment (immortal person-time).² The authors do not describe any analytical measures taken to counteract this conditioning. If they have not taken this into account, this would pose a very serious problem as also described numerous times in the epidemiological literature.³ (Hanley JA, 2014) Immortal time bias will generate an illusion of treatment effectiveness and is frequently found in observational studies that compare with non-users.⁴ (Suissa S, 2007)

In general, suspicion for immortal time bias should be raised when exposure groups are assigned with no regard to exposure time in a longitudinal study. Furthermore, the reported effect sizes are surprisingly high, especially when considering the fragility of the population in question. This group consists of stroke patients with complications; they may suffer from e.g. post-stroke delirium and may need antipsychotic treatment. They would most likely have more adverse outcomes, not the opposite.^5,6 (Shi Q, 2012, Prior A, 2014)

Dr. Anders Prior, MD

Research Unit for General Practice and Section for General Medical Practice, Department of Public Health, Aarhus University, Denmark

Dr. Thomas Munk Laursen, PhD

National Centre for Register-based Research, Department of Economics and Business, Aarhus University, Denmark

Prof. Mogens Vestergaard, PhD

Research Unit for General Practice and Section for General Medical Practice, Department of Public Health, Aarhus University, Denmark

References

1 Wang JY, Wang CY, Tan CH, Chao TT, Huang YS, Lee CC. Effect of different antipsychotic drugs on short-term mortality in stroke patients. Medicine (Baltimore). 2014;93(25):e170.

2 Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 3rd ed. Philadelphia, PA: Lippincott Williams & Wilkins; 2008.

3 Hanley JA, Foster BJ. Avoiding blunders involving 'immortal time'. Int J Epidemiol. 2014;43(3):949-961.

4 Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf. 2007;16(3):241-249.

5 Shi Q, Presutti R, Selchen D, Saposnik G. Delirium in acute stroke: A systematic review and meta-analysis. Stroke. 2012;43(3):645-649.

6 Prior A, Laursen TM, Larsen KK, et al. Post-stroke mortality, stroke severity, and preadmission antipsychotic medicine use--a population-based cohort study. PLoS One. 2014;9(1):e84103.