New use of prescription drugs prior to a cancer diagnosis

Pharmacoepidemiol Drug Saf. 2017 Feb;26(2):223-227. doi: 10.1002/pds.4145. Epub 2016 Nov 27.


Purpose: Cancers often have considerable induction periods. This confers a risk of reverse causation bias in studies of cancer risk associated with drug use, as early symptoms of a yet undiagnosed cancer might lead to drug treatment in the period leading up to the diagnosis. This bias can be alleviated by disregarding exposure for some time before the cancer diagnosis (lag time). We aimed at assessing the duration of lag time needed to avoid reverse causation bias.

Methods: We identified all Danish patients with incident cancer between 2000 and 2012 (n = 353 087). Incident use of prescription drugs was assessed prior to their cancer diagnosis as well as among population controls (n = 1 402 400). Analyses were conducted for all cancers and for breast, lung, colon and prostate cancer individually. Further, analyses were performed for a composite measure of all incident drug use as well as for nine pre-specified individual drug classes, representing drug treatment likely to be prescribed for symptoms of the given cancers.

Results: The incidence rate for new drug treatment among cancer cases was stable around 130 per 1000 persons per month until 6 months prior to cancer diagnosis where it increased gradually and peaked at 434 in the month immediately preceding the cancer diagnosis. Considerable variation was observed among cancers, for example, breast cancer showed almost no such effect. The pre-selected drug classes showed a stronger increase prior to cancer diagnoses than drugs overall.

Conclusions: Incident use of drugs increases in the months prior to a cancer diagnosis. To avoid reverse causation, 6 months' lag time would be sufficient for most drug-cancer associations. © 2016 The Authors. Pharmacoepidemiology and Drug Safety published by John Wiley & Sons Ltd.

Keywords: adverse drug effects; cancer; epidemiology; pharmacoepidemiology; reverse causation bias.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Bias*
  • Denmark
  • Female
  • Humans
  • Incidence
  • Male
  • Neoplasms / diagnosis*
  • Neoplasms / etiology
  • Neoplasms / pathology
  • Pharmacoepidemiology / methods*
  • Prescription Drugs / administration & dosage*
  • Prescription Drugs / adverse effects
  • Registries
  • Time Factors


  • Prescription Drugs