Policy implications of adjusting randomized trial data for economic evaluations: a demonstration from the ASCUS-LSIL Triage Study

Abstract

Background: Although the randomized controlled trial (RCT) is widely considered the most reliable method for evaluation of health care interventions, challenges to both internal and external validity exist. Thus, the efficacy of an intervention in a trial setting does not necessarily represent the real-world performance that decision makers seek to inform comparative effectiveness studies and economic evaluations.

Methods: Using data from the ASCUS-LSIL Triage Study (ALTS), we performed a simplified economic evaluation of age-based management strategies to detect cervical intraepithelial neoplasia grade 3 (CIN3) among women who were referred to the study with low-grade squamous intraepithelial lesions (LSIL). We used data from the trial itself to adjust for 1) potential lead time bias and random error that led to variation in the observed prevalence of CIN3 by study arm and 2) potential ascertainment bias among providers in the most aggressive management arm.

Results: We found that using unadjusted RCT data may result in counterintuitive cost-effectiveness results when random error and/or bias are present. Following adjustment, the rank order of management strategies changed for 2 of the 3 age groups we considered.

Conclusions: Decision analysts need to examine study design, available trial data, and cost-effectiveness results closely in order to detect evidence of potential bias. Adjustment for random error and bias in RCTs may yield different policy conclusions relative to unadjusted trial data.

Figure 1. Natural history of cervical carcinogenesis

Schiffman M, Kjaer SK. Natural history of anogenital human papillomavirus infection and neoplasia. J Natl Cancer Inst Monogr 2003;31:14–19, by permission of Oxford University Press. This figure depicts the natural history of cervical carcinogenesis. HPV incidence peaks around the time of sexual debut, and while most infections are benign and clear within 1 to 2 years of sexual transmission, some will persist. Persistent infections with any of the 10 to 15 carcinogenic HPV types are at risk of developing into precancer. Histologically confirmed CIN3 is at high risk of progressing to cancer and requires definitive treatment. HPV: human papillomavirus; CIN: cervical intraepithelial neoplasia.

Figure 2. Model overview

This figure displays the general structure of the decision tree model used to evaluate costs and number of cases of CIN3 detected among women with a referral cytology of LSIL in ALTS. Management strategies included immediate colposcopy (IC), HPV DNA testing, and conservative management (CM) with up to three repeat cytology tests at enrollment, 6 months, and 12 months. Model health state was based on the worst 2-year QC pathology group diagnosis. Triage test results were only relevant in the HPV DNA testing arm (Hybrid Capture 2 [HC2]) and CM (cytology) arms, and women were referred to colposcopy for a positive or missing HC2 test, or HSIL+, respectively. Colposcopy results were based on evaluation at clinical centers, with a treatment threshold of CIN2. The outcome was cost per case of CIN3 detected. CIN: cervical intraepithelial neoplasia; LSIL: low-grade intraepithelial leson; ALTS: ASCUS/LSIL Triage Study; IC: immediate colposcopy; HPV: human papillomavirus; CM: conservative management; HC2: Hybrid Capture 2; QC: quality control; HSIL: high-grade squamous intraepithelial lesion.