The Northern Ireland local group of the RSS held an online meeting using MS Teams on Wednesday 4 May 2022 at 2pm (BST).
The speaker was Professor Gilbert MacKenzie, formerly of the Centre of Bio-statistics, University of Limerick, Ireland.
Professor MacKenzie described the Northern Ireland Lung Cancer Study (NILCS), the first population-based, multi-source, study of the incidence of lung cancer conducted between 1991 and 1992. The study was not only of historical interest, but the original findings still resonate today. The study was unique in that ascertainment was completed by one dedicated medical worker, Dr Pauline Wilkinson.
In 1991 there was no Cancer Registration in Northern Ireland which relied on a Notification System thought to be incomplete and several agencies had switched to using the lung cancer death rate as a surrogate for incidence. When NILCS was completed, the Notification System was found to be 40.1% complete and the Death Certification was 66.7% complete. No single source held all 900 cases ascertained. Professor MacKenzie showed the incidence in NI and in the British Isles: the Republic of Ireland had the lowest rate, 40.8 (per 100,000 pop.), followed by NI (57.0), England and Wales (71.4) and Scotland (97.1). The male to female ratio was 2:1 in all regions except Scotland, where the rate among females was higher. The most striking findings lay in the trends from 1991-2018 which showed a general pattern of increasing incidence rates throughout the British Isles, with a particularly sharp increase in NI since 2002.
Gilbert investigated the survival of these 900 incident cases. The median survival was only 4.7 months and the one year survival was 30%. A telling finding was that 51.6% of patients received palliative care. Surgery was the best treatment, followed by combined therapy (chemotherapy and radiotherapy), chemotherapy alone and radiotherapy alone in rank order. But the beneficial effect of surgery was exaggerated in these observational data as those receiving surgery were highly selected with advantageous profiles in the other covariates. Treatment survival did not follow a proportional hazards model and a non-PH MPR Weibull model (Burke & MacKenzie, 2017), which modelled the scale and shape parameters with separate linear predictors fitted the data well. Some 9 covariates were modelled and survival was found not to depend on age nor sex. The effect of adjustment for other relevant covariates was to reduce the benefit attributable to surgical treatment by more than half.
Selection for treatment was considered next. The survival covariates were used in a logistic model for the probability of receiving curative therapy. The model showed heavy dependence on several factors including age? A screening table was produced which showed that 105 patients which the model predicted should be treated did not receive curative treatment (Group A), and that an- other 105 patients were treated when the model predicted they should not be treated (Group B). Survival in this latter group was improved (over palliative care) and Professor MacKenzie suggested that a similar or better improvement might have occurred in Group A had they been treated in accordance with the prediction. He went on to analyse the difference in survival between Group A and those who had been treated in accordance with the model, showing that the difference was real, and that survival might be improved upon if it were possible to treat, at least some, patients in Group A.
Professor MacKenzie concluded by saying that the increasing trends in incidence were worrying and that further research was required to discover the reasons. He felt that these trends had been rather overlooked and deplored the lack of progress on the survival front, suggesting the formation of a task force to address these key issues. On selection for treatment. he pointed to the need to create relevant and consistent treatment models in this era of personalised medicine.
The talk was well received and the audience thanked the speaker in the usual way.
The speaker was asked about the possible inclusion of time dependent functions in the shape parameter of the MPR model. Prof. MacKenzie said the shape parameter was already performing that role (viz: the MPR Survival curves) and so this should not be necessary. He was also asked about relaxing the dichotomy of the treatment score and agreed that this was entirely feasible.
Hannah Mitchell, Chair, closed the meeting at 15:10 saying that our next meeting would be held in the Autumn term.
Reference: Burke, K. and MacKenzie, G. (2017). Multi-parameter regression survival modelling - an alternative to proportional hazards, Biometrics, Vol. 73:361-716.
Written by Gilbert MacKenzie.