When assumptions go awry: practical alternatives for modeling the mortality reductions in cancer screening
Bibliographic record
Abstract
Background and research hypothesis: The aim of cancer screening (CS) is to detectcancers at earlier stages before the individual presents any symptoms, in the hope of deliveringearlier treatment, reducing the cost of treatment, and increasing the chances of survival. Screeningfor cervical, breast, colon, prostate, and lung cancer has become part of regular medical care inmany countries including Canada. To analyze data from cancer screening trials, and estimate themortality reductions, conventionally the Cox proportional hazards (Cox PH) model is used. Thismodel is powerful in situations where the effect of the intervention is immediate and sustained, aswith vaccines. However, this immediacy is not the case for CS: the assumption of proportionalityof hazards between the screened and the unscreened arms, i.e, of a constant hazard ratio, is notrealistic. We hypothesize that continuing to use this assumption leads to biased estimates of themortality reductions that can be expected from cancer screening programs. Therefore, we explorealternate methods, such as the Royston-Parmar model, to minimize the bias and to produce nonconstant-in-time hazard ratios.Research objective: The aims are to quantify the impact of violations of the proportionalhazards assumption in the Cox PH model on the estimation of the rate (hazard) ratio function. Wedo so using simulated data and real data from the National Lung Screening trial (NLST). Wepropose practical alternatives.Method and analysis: Using the statistical software R, we: 1) Simulated data from theCox PH model, tested the different analysis models and compared the results and interpretations.2) Simulated data from the Royston-Parmar model, tested the different models and compared them.This allowed us to see the impact on effect parameter estimation when the proportional hazardassumptions are violated. 3) Simulated data from a time-dependent Cox model and tested andcompared the different models. This was a second situation when the proportional hazardassumptions are violated. We then compared the results of each model (Cox PH model, Royston-Parmar and time-dependent Cox model) depending on each of the three data generating scenarios.This allowed one to see the effects of assumption violations, as well as to understand how eachmodel behaves depending on the underlying data-generating mechanism. Finally, we use the datafrom the National Lung Screening Trial (NLST), a randomized trial, to test the three models andsee their behaviours in a real setting.Results and Conclusion: The empirical comparison showed that the Royston-Parmarmodel performed well for data with a time varying hazard ratio than the Cox PH model. However,with a higher complexity of the underlying time varying model, the performance of the Royston-Parmar model was worse than the extended Cox model. With the NLST data, both Royston-Parmarand the extended Cox models showed that the ratio of the rates of death from lung cancer and oflung cancer diagnosis at stage III and IV were not constant over follow-up time
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.077 | 0.202 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.003 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.004 |
| Scholarly communication | 0.004 | 0.006 |
| Open science | 0.005 | 0.004 |
| Research integrity | 0.006 | 0.007 |
| Insufficient payload (model declined to judge) | 0.008 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".