Vernon T. Farewell

Background information is provided for Peter Armitage’s recollections of the early years of the Medical Research Council Statistical Research Unit.

Objective: To determine predictors of change in neuropsychiatric (NP) event status in a large, prospective, international inception cohort of patients with systemic lupus erythematosus (SLE).Methods: Upon enrollment and annually thereafter, NP events attributed to SLE and non‐SLE causes and physician‐determined resolution were documented. Factors potentially associated with the onset and resolution of NP events were determined by time‐to‐event analysis using a multistate modeling structure.Results: NP events occurred in 955 (52.3%) of 1,827 patients, and 593 (31.0%) of 1,910 unique events were attributed to SLE. For SLE‐associated NP (SLE NP) events, multivariate analysis revealed a positive association with male sex (P = 0.028), concurrent non‐SLE NP events excluding headache (P < 0.001), active SLE (P = 0.012), and glucocorticoid use (P = 0.008). There was a negative association with Asian race (P = 0.002), postsecondary education (P = 0.001), and treatment with immunosuppressive drugs (P = 0.019) or antimalarial drugs (P = 0.056). For non‐SLE NP events excluding headache, there was a positive association with concurrent SLE NP events (P < 0.001) and a negative association with African race (P = 0.012) and Asian race (P < 0.001). NP events attributed to SLE had a higher resolution rate than non‐SLE NP events, with the exception of headache, which had comparable resolution rates. For SLE NP events, multivariate analysis revealed that resolution was more common in patients of Asian race (P = 0.006) and for central/focal NP events (P < 0.001). For non‐SLE NP events, resolution was more common in patients of African race (P = 0.017) and less common in patients who were older at SLE diagnosis (P < 0.001).Conclusion: In a large and long‐term study of the occurrence and resolution of NP events in SLE, we identified subgroups with better and worse prognosis. The course of NP events differs greatly depending on their nature and attribution.

Plaid designs are characterised by having one set of treatments applied to rows and another set of treatments applied to columns. In a 2003 publication, Farewell and Herzberg presented an analysis of variance structure for such designs. They presented an example of a study in which medical practitioners, trained in different ways, evaluated a series of videos of patients obtained under a variety of conditions. However, their analysis did not take full account of all error terms. In this paper, a more comprehensive analysis of this study is presented, informed by the recognition that the study can also be regarded as a two-phase design. The development of random effects models is outlined and the potential importance of block-treatment interactions is highlighted. The use of a variety of techniques is shown to lead to a better understanding of the study. Examination of the variance components involved in the expected mean squares is demonstrated to have particular value in identifying appropriate error terms for F-tests derived from an analysis of variance table. A package such as ASReml can also be used provided an appropriate error structure is specified. The methods presented can be applied to the design and analysis of other complex studies in which participants supply multiple measurements under a variety of conditions.

For rheumatic diseases, Minimal Disease Activity (MDA) is usually defined as a composite outcome which is a function of several individual outcomes describing symptoms or quality of life. There is ever increasing interest in MDA but relatively little has been done to characterise the pattern of MDA over time. Motivated by the aim of improving the modelling of MDA in psoriatic arthritis, the use of a two-state model to estimate characteristics of the MDA process is illustrated when there is particular interest in prolonged periods of MDA. Because not all outcomes necessary to define MDA are measured at all clinic visits, a partially hidden multi-state model with latent states is used. The defining outcomes are modelled as conditionally independent given these latent states, enabling information from all visits, even those with missing data on some variables, to be used. Data from the Toronto Psoriatic Arthritis Clinic are analysed to demonstrate improvements in accuracy and precision from the inclusion of data from visits with incomplete information on MDA. An additional benefit of this model is that it can be extended to incorporate explanatory variables, which allows process characteristics to be compared between groups. In the example, the effect of explanatory variables, modelled through the use of relative risks, is also summarised in a potentially more clinically meaningful manner by comparing times in states, and probabilities of visiting states, between patient groups.