Archives

  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • 2021-03
  • br Randomisation and masking br Two types of randomisations

    2019-09-23


    2.2. Randomisation and masking
    Two types of randomisations were used: randomisation before consent (Zelen-type effectiveness design in Sweden, Finland, and Italy) and randomisation after consent (efficacy design in the other countries). Randomisation was done by computer-generated random numbers, with eligible participants identified in Kainic acid registers. Trial group allocation was masked for determination of the main outcome.
    The primary outcome was PCa mortality. For deceased men with PCa, medical records were evaluated by a cause of death (COD) committee using a standardised flow-chart to establish the COD [7]. The COD committee was masked regarding the randomisation arm. Official CODs were used in Finland since 2003 after demonstrating a very high concordance with that obtained by the local COD committee [8]. PCa
    incidence[4T$DIF]and vital status were[13T$DIF] monitored[14T$DIF] regularly[15T$DIF] in[16T$DIF] all[17T$DIF] randomised[18T$DIF] [19T$DIF] men and reported biannually to the central database. For men with PCa, TNM stage, PSA, Gleason score, and primary treatment were abstracted from medical records. A scientific committee established quality criteria and other committees monitored the conduct, progress of the trial, PSA harmonisation, and assignment of Gleason grades [5]. This report includes follow-up through December 31, 2014 or a maximum of 16 yr after randomisation.
    2.4. Statistical analysis
    The primary analysis evaluated PCa mortality and focused on the core age group of men 55–69–yr old, with follow-up through 2014 truncated at 9, 11, 13, and 16 yr. The main analysis was carried out according to the intention-to-screen principle, that is, comparing groups formed by randomisation (regardless of assignment compliance). Incidence and mortality rates, and risks were calculated by dividing the number of events by the number of person years and the number of men, respectively. Rate ratios (RRs; ratio of incidence per person year), risk ratios (ratio of incidence per man), and the corresponding differences were calculated using Poisson regression analysis, with the control arm for Finland weighted by 1:1.5 due to unequal allocation (agreed upon when Finland joined the trial). Confidence intervals (CIs) for rate and risk differences were calculated by Wald’s method, with standard errors derived by the delta method. For the number needed to invite (NNI), the CIs were derived as 1 over the intervals for the differences in the risk of PCa mortality. The p values are two sided. No adjustment of significance for alpha-spending in sequential analyses was applied because the present analysis is protocol based and not driven by statistical significance [9,10]. The NNI to avert one PCa death was calculated as the inverse of the absolute risk difference in PCa deaths. The number needed to detect (NND) was defined as the NNI multiplied by the excess incidence of PCa in the screening group. Both the graphs on cumulative PCa incidence and mortality in the control and screening arms, and the graph on survival after screen-detected PCa are based on Nelson-Aalen estimates of survival.]FID$T6[Cumulative incidence and mortality curves adjusted for the competing risk of death of other causes follow the approach described in the study of Kalbfleisch and Prentice [11].
    In a secondary analysis, PCa mortality was assessed from diagnosis in those men diagnosed within the programme. Men with screen-detected 
    cancer in round 1 were compared with screen-detected men during subsequent screening rounds. Cox regression analysis was used.
    To evaluate the effect of attending at least one screening round, adjusted RRs were calculated with adjustment for nonparticipation [12]. The proportion of complete nonattendees (ie, never participating) in the screening group and the PCa mortality among them were calculated. The control group is then considered to consist of a nonattender part of the same size and the same PCa mortality rate as the nonattender part of the screening group, allowing us to calculate the adjusted mortality rate among those hair bulb participated at least once.