Many studies have sought to answer the question Does acetaminophen (paracetamol) cause cancer?, but these studies show inconsistent results. One reason for this inconsistency could be that the study designs that were used are prone to bias. Here we aim to evaluate these designs by using a large set of negative control outcomes, things we know are not caused by acetaminophen, and determine how often the various study designs used in the past get the right answer for these negative controls (in relationship to acetaminophen).
We will evaluate two main types of study designs, so-called case-control designs, and cohort designs. Well test several variants of each. The various design choices are all based on the designs used in previously published studies.
We will execute these designs against the CPRD database to produce estimates for the negative controls, as well as several types of cancer. This will allow us to see how far off the results for the negative controls are from the truth (that there is no effect), as well as how far away the results for the cancers are from the negative controls. The results of this study could very well help explain the inconsistencies between the various studies.
A large number of epidemiologic studies have been conducted to examine whether use of acetaminophen predisposes to the occurrence of one or more forms of cancer. There are many limitations to many of these studies as noted earlier, including vulnerability to channeling, protopathic bias, and uncontrolled confounding. However, the magnitude of the bias resulting from these limitation remains unknown, hampering the interpretability of the results of these studies. Recent methodological developments have focused on using large sets of negative controls exposure-outcome pairs where no causal effect is believed to exist to measure the operating characteristics of study designs by observing to what extent these designs produce effect size estimates in line with the truth (that there is no effect for the negative controls).
Here we aim to emulate prior studies, while including negative controls to quantify residual bias in these study designs. These prior studies mostly followed a case-control designs, although some used a cohort design. We will mimic the design choices in these prior studies as best we can, including the mechanism by which controls were selected, how exposure was defined, as well as the covariates used to adjust for potential confounding. We define 8 variants of the case-control design, and 2 variants of the cohort design.
The 37 negative controls were used in a previous study,2 and were selected based on a lack of evidence in literature, product labels, and spontaneous reports, as well as a manual review by several clinicians. In addition to the negative controls we also include four cancer outcomes.
The negative controls will allow quantification of the error due to the limitations of these study designs. This quantification in turn can be used to help interpret study results by determining whether an observed effect size falls outside of what can be expected based solely on error (both systematic and random error).
37 negative control outcomes: Achilles tendinitis; Atrophic vaginitis; Breath smells unpleasant; Bronchiectasis;
Disorders of initiating and maintaining sleep; Ear problem; Erythema nodosum; Falls; Foot-drop; Ganglion and cyst of synovium, tendon and bursa; Hemangioma; Hydrocele; Hyperthyroidism; Impaired glucose tolerance; Impingement syndrome of shoulder region; Impotence; Incontinence of feces; Interpersonal relationship finding; Irregular periods; Irritability and anger; Joint stiffness; Loss of sense of smell; Mixed hyperlipidemia; Osteitis deformans; Panic attack; Perforation of tympanic membrane; Pes planus; Polymyalgia rheumatica; Premature menopause; Prolapse of female genital organs; Pure hypercholesterolemia; Respiratory symptom; Restless legs; Restlessness and agitation; Rosacea; Simple goiter; Skin sensation disturbance; Snapping thumb syndrome; Urinary symptoms
4 outcomes of interest: Renal cell carcinoma; Primary liver cancer; Lymphoma; Multiple myeloma
Martijn Schuemie - Chief Investigator - Janssen US
Martijn Schuemie - Corresponding Applicant - Janssen US
Patrick Ryan - Collaborator - Janssen Research & Development LLC