Study questions Flashcards
Why not use a Comorbidity index? such as charlson?
Charlson Comorbidity Index was developed to predict death in one year, not cancer.
We adjusted for major comorbidities, one of which (joint replacement surgery) has been identied as a predictor of cancer. Plus we adjsuted for hospitalizations and drugs, and so we have pretty much the same info as in an index.
OTC NSAID
Not captured, but use of over the counter NSAIDs in this population is low, (not reimbursed)
Much more cancer in rituximab cohort, problem?
We restricted the analysis to first cancer so should not give rise to selection bias.
Why did you group the cancers in this way?
- Site specific models needed
- Power
- Previous signals
What is the accuracy of the swedish cancer register?
The completeness in 1998 was good. Only 3,7% of cancers were missed, and the numbers were even better for breast and female genital cancer.
What is the accuracy of the swedish patient register?
The inpatient register has been validated and found to be good. For RA, the PPV was 90-95%, and sensitivity 42%
I don’t know of any publication on the outpatient register as a whole, but I know that for underreporting is presumably low, and based on some private practiotioners and primary care RA (although the accuracy of these diagnoses is unclear).
9/260 Swedish rheumatologists were private practiotioners, and from what i’ve seen they do report to the patient register and ARTIS, but maybe to a lwer extent.
Primary care RA has been estimated by someone to be up to 10%, but it’s unclear what type of patients these really are.
Liknande studiedesign studie 1-3 men lite olika vad gäller detaljerna
- PhD is a learning process, i’m not saying that the studies get better and better but my perception of things change.
- SOmetimes a study specific rationale
- Some degree of arbitrariness
Hur funkar cancerregistret?
1 observation per tumör 2. Incidenta cancrar. 3 Status vid diagnos 4. Dubbelrapportering 5. Mandatory at reportera för alla health care providers
Time varyig covariates comorbidites
Trade off. A conservative approach would be to only use index date status, but then we would underestimate true comorbidites. If we include information during follow up, then we might adjust for mediators, which is not good, depending on the covariate and the outcome we can speculate on how likely this could be.
Restriciting to one year before diagnosis is because we dont want to risk it being a symptom of the outcome.
Why not put all (bionaive, TNFi and genpop) in the same model.
- By separating them we can more easily say, what seems to be the disease, and what seems to be the exposure.
- Covariates might mean different things for genpop and RA, if we put all in the same (with 10 times more genpop than RA) then we will force the value of the covariates in the genpop on the RA population. For example, hospitalizations before index might meen different things within RA comparisons (ra disease severity, complciations, comorbidites) than in genpop)
Patients in more than one cohort, important?
Important that person time is accurately counted towards the correct exposure group. You might have to handle this in your regression model as well, otherwise the confidence intervals might be falsely narrow.
First or second bdmard, why important.
Why 1st Tnfi?
Why 2nd Tnfi?
If not first bmdard, then the patient has failed on a previous biologic which might be associated with the outcome.
1St TNfi is a common defintion in other studies and using this therefore enhances comparability to those studies. Also by using 1st TNFi in bionaive we ensure that otehr potential risks by other bdmards doesent mix up the results (less important for the other bdmards because more evidence of no increased risk of cancer with TNFi)
2nd TNFi is because most patients on nTNFi have failed on TNFi so more relevant comparison
Maybe a better comparison would be regardless of previous bDMARD but adjusted for number of previous bDMARDs
Smoking
We handled this bla bla.
Similar between TNFi and csDMARD-treated.
Bias quantification analysis in study 2.
Also, i’m not too sure how well I trust self-reported smoking status, I think that with an increasing stigma associated with smoking over time, it might be even worse nowadays.
Also what to adjust for, current smoking, ever smoking. What if starting TNFi treatment is associated with quitting smoking, or if treatment is reserved for those that dont smoke.
Aspirin confounder? Why not adjust
So, a lot of observational studies have shown protective effects, and some RCTs as well. But a huge RCT (40,000) women (Women’s Health Study (WHS)) showed absolutely no relation between low-dose aspirin and risk of breast cancer. Observational studies have shown a 25% risk reduction but are heterogenous, and the largest risk reductions were seen in those with low-dose aspirin.
So the proposed effects of aspirin take a lot of time to kick in, and although I could see aspirin use being higher after diagnosis, and some time before, I don’t reallty think that it would explain a reduction more than 10 years before RA diagnosis.
Furthermore adjusting for NSAIDs in our study would be hard. First of all, you would expect that NSAID-use means different things in RA-patients than in the general population (joint pain, hear condition etc. ). And we would presumably capture this differently, with RA-patients less likely to use over the counter drugs since they are not reimbursed.
Validity of RA diagnosis
PPV of 91%, when validated against ACR criteria, Those that did not have RA had another rheumatic disease.
Higher PPV for seropositive.
Sometimes there is difficult overlap, for example erosive arthritis in a ACPA+ positive patient with cutaneous psoriasis
