Regression calibration of self-reported mobile phone use to optimize quantitative risk estimation in the COSMOS study

Authors: Reedijk M, Portengen L, Auvinen A, Kojo K, Heinävaara S, Feychting M, Tettamanti G, Hillert L, Elliott P, Toledano MB, Smith RB, Heller J, Schüz J, Deltour I, Poulsen AH, Johansen C, Verheij R, Peeters P, Rookus M, Traini E, Huss A, Kromhout H, Vermeulen R, Study Group TC

Year: 2024 May 13

Category: Epidemiology

Journal: American Journal of Epidemiology

DOI: 10.1093/aje/kwae039

URL: https://academic.oup.com/aje/advance-article/doi/10.1093/aje/kwae039/7671112?login=false

Abstract

Overview

The Cohort Study of Mobile Phone Use and Health (COSMOS) has repeatedly collected self-reported and operator-recorded data on mobile phone use. Assessing health effects using self-reported information is prone to measurement error; however, operator data were available prospectively for only part of the study population and did not cover past mobile phone use.

Methods

To optimize data and reduce bias, different statistical approaches for constructing mobile phone exposure histories within COSMOS were evaluated. The study compared four regression calibration (RC) methods (simple, direct, inverse, and generalized additive model for location, shape, and scale), complete-case (CC) analysis, and multiple imputation (MI) in a simulation study with a binary health outcome. Data consisted of self-reported and operator-recorded mobile phone call data collected at baseline (2007-2012) from participants in Denmark, Finland, the Netherlands, Sweden, and the UK.

  • Self-reported duration of mobile phone use was based on weekly usage categories.
  • Operator-recorded duration was used when available and covered all phones reported by participants.
  • Recorded call duration was analyzed by country, showing variation in geometric mean values across populations.

Findings

Parameter estimates from simple, direct, and inverse RC methods were associated with less bias and lower mean squared error than those from CC analysis or MI. RC methods yielded more accurate estimations of the relationship between mobile phone use and health outcomes by combining self-reported with objective operator-recorded data available for a subset of participants.

Exposure Measurement and EMF Relevance

A major issue outlined is the ability of mobile phone use to predict the exposure of interest: radiofrequency electromagnetic fields (RF-EMF). Past validation studies for earlier mobile phone generations supported the link between phone use and RF-EMF exposure; however, less predictive accuracy is documented for 3rd, 4th, and 5th generation devices currently or recently used by study participants. This underlines the consistent concern regarding EMF exposure and possible health effects, justifying improved exposure assessment methods for robust risk estimation.

Conclusion

This study addresses a significant concern in mobile phone research and environmental epidemiology: improving exposure-outcome relation estimation by combining error-prone self-reports with more objective, but less available, operator records. RC approaches appear to improve exposure-outcome estimation and, through COSMOS' prospective design, will allow more robust conclusions about possible health effects from mobile phone use and related EMF exposure.

mobile phone use regression calibration electromagnetic fields exposure assessment risk estimation COSMOS study health outcomes
← Back to Stats