Charlotte Davies1, Andrew Briggs2, Paula Lorgelly3, Göran Garellick4, Henrik Malchau5. 1. Health Economics Group, Norwich Medical School, University of East Anglia, United Kingdom (CD) 2. Section of Public Health and Health Policy, University of Glasgow, United Kingdom (AB) 3. Centre for Health Economics, Monash University, Australia (PL) 4. Department of Orthopaedics, Sahlgrenska University Hospital, Sweden, Swedish Hip Arthroplasty Register, Go¨ teborg, Sweden (GG) 5. Harvard Medical School, Boston, Massachusetts, Swedish Hip Arthroplasty Register, Go¨ teborg, Sweden (HM)
Abstract
BACKGROUND: It is widely recommended that health technology appraisals adopt a lifetime horizon to assess the relative costs and benefits of an intervention. However, most trials or clinical studies have relatively short follow-up periods, with the event of interest not occurring before the end of the study for many subjects. In such cases, survival analysis using parametric models can be used to extrapolate into the future. OBJECTIVE: To assess the accuracy of survival analysis in projecting future events beyond the sample estimation period. DESIGN: Using a previously published comparison of 2 alternative hip replacement prostheses based on 8 years of data as a case study, we extend the data set to include 8 years more data. Using the new data, the parametric assumptions of the previous study and its success in predicting the outcomes are assessed. RESULTS: The extended data set casts doubt on the previous study's findings. The failure curves of the 2 prostheses now cross, and the proportional hazards assumption no longer holds. Extrapolations from the original data set yielded very good predictions for one prosthesis for the full 16 years but were much poorer for the other, even when the proportionality assumption was relaxed. CONCLUSIONS: Care should be taken when extrapolating treatment benefits for new technologies early in their life cycle based on observational or randomized controlled trial data sources. This case study reveals that predictions of prosthesis failure based on a short follow-up period were inaccurate compared with those after a longer period of follow-up.
BACKGROUND: It is widely recommended that health technology appraisals adopt a lifetime horizon to assess the relative costs and benefits of an intervention. However, most trials or clinical studies have relatively short follow-up periods, with the event of interest not occurring before the end of the study for many subjects. In such cases, survival analysis using parametric models can be used to extrapolate into the future. OBJECTIVE: To assess the accuracy of survival analysis in projecting future events beyond the sample estimation period. DESIGN: Using a previously published comparison of 2 alternative hip replacement prostheses based on 8 years of data as a case study, we extend the data set to include 8 years more data. Using the new data, the parametric assumptions of the previous study and its success in predicting the outcomes are assessed. RESULTS: The extended data set casts doubt on the previous study's findings. The failure curves of the 2 prostheses now cross, and the proportional hazards assumption no longer holds. Extrapolations from the original data set yielded very good predictions for one prosthesis for the full 16 years but were much poorer for the other, even when the proportionality assumption was relaxed. CONCLUSIONS: Care should be taken when extrapolating treatment benefits for new technologies early in their life cycle based on observational or randomized controlled trial data sources. This case study reveals that predictions of prosthesis failure based on a short follow-up period were inaccurate compared with those after a longer period of follow-up.
Authors: Brandon K Bellows; Yiyi Zhang; Zugui Zhang; Donald M Lloyd-Jones; Adam P Bress; Jordan B King; Paul Kolm; William C Cushman; Karen C Johnson; Leonardo Tamariz; Elizabeth C Oelsner; Steven Shea; Anne B Newman; Diane G Ives; David Couper; Andrew E Moran; William S Weintraub Journal: J Am Heart Assoc Date: 2021-05-06 Impact factor: 5.501
Authors: Elsa M R Marques; Rachel Humphriss; Nicky J Welton; Julian P T Higgins; William Hollingworth; Jose A Lopez-Lopez; Howard Thom; Linda P Hunt; Ashley W Blom; Andrew D Beswick Journal: Syst Rev Date: 2016-02-01
Authors: Lasse H Jakobsen; Therese M-L Andersson; Jorne L Biccler; Tarec C El-Galaly; Martin Bøgsted Journal: BMC Med Res Methodol Date: 2019-01-28 Impact factor: 4.615