OHDSI Home | Forums | Wiki | Github

"Seek COVER" PLP study posted to MedRxiv ... please share!

General
#1

As mentioned in our earlier update, the study Seek COVER: Development and validation of a personalized risk calculator for COVID-19 outcomes in an international network is designed to inform individual behavioral choices and help design shielding strategies during de-confinement. This has just been posted to MedRxiv, so please share this link with your respective networks!

Check out the graphic below the link. As described in the study, this shows “a risk converter, which allows for easy conversion between the risk score and predicted risk of the outcomes. Furthermore, we provide a plot of the probability distribution for the three models from patients in ClinFormatics to demonstrate the expected regions the probabilities fall into. To calculate the COVER scores using Figure 1, a clinician needs to identify which predictors the patient has. The points for each of those predictors are then added to arrive at the total score. For example, if a 63-year-old female patient has diabetes and heart disease, then her risk score for hospital admission (COVER-H) is 43 (female sex) + 4 (heart disease) + 3 (diabetes) + 15 (age) = 65. The risk scores for intensive services (COVER-I) and fatality (COVER-F) are 51 and 47, respectively.”

Led by co-first authors @RossW and @aniekmarkus, the team designed a nine-predictor COVID-19 Estimated Risk (COVER) model that was validated using more than 43,000 COVID patients (following initial development and validation using more than 6.8 million patients with influenza or flu-like symptoms). This model predicts hospitalization, intensive services, and death, and can help provide reassurance for low-risk patients, while shielding high-risk patients, as many start to enter the de-confinement stage of the pandemic.

@Rijnbeek is the corresponding author, and both he and @jennareps are co-last authors. Overall there are 43 authors involved in the study, once again highlighting the global collaborative nature of the OHDSI community. Congratulations to all who were involved in this work.

24%20PM
24%20PM2240×1550 607 KB

#2

Congratulations on this publication.

At the moment, the listed link to the Shiny app in the paper appears to be broken:
http://evidence.ohdsi.org/Covid19CoverPrediction

(there is a period in the hyperlink in the pdf, but it also doesn’t work when you remove it)

Is this the working link or is it out of date?
https://data.ohdsi.org/Covid19CoverPrediction/

Thanks,
Christophe

#3

I’m curious what the authors make of hyperlipidemia being protective in the model. The paper does not comment on this point, and it is counterintuitive.

#4

Hi Christophe,

thanks for the interest in the paper the evidence viewer is now available!

Regarding hyperlipidemia being “protective”. I would not move to a causal interpretation of the model coefficients based upon this research. We created a multivariate model so the coefficients should only be interpreted in the context of their function with each other. What i think we could be seeing is that the hyperlipidemia is correcting for an overestimate of effect in the other covariates and as such it receives a negative coefficient.

Could be a nice study to investigate it though!

#5

But in terms of clinical plausibility, what do we know about hyperlipidaemia patients? Were they treated with e.g. statins (some suggest these are protective) or indeed treated with ACEi/ARBs?

Are they homogenous or heterogenous?

If the prediction model could be used for clinical prediction, and of course correlation is not causation, but clinical plausibility needs also to be evaluated to support the findings?

#6

Thanks for fixing the link. I was not attributing causality – a risk or protective factor can be an association. In any event, hyperlipidemia was the largest magnitude coefficient besides age impacting fatality risk, and the only one associated with better outcomes besides youth. When single factors are examined for COVID-19 in epidemiology studies, hyperlipidemia is associated with worse outcomes – but that could be confounded, among other things, with age and obesity. You did not include the latter in your model, making the results more surprising. It cut my risk of death in half in the model to check the hyperlipidemia box. I agree with @nigehughes that some attempt at clinical explanation is warranted.

#7

Thanks for this interesting discussion. The potential protective effect of hyperlipidemia could be due to an indirect association with statins, which have appeared as protective in other manuscripts, too, but these effects seem to be independent, given that models were adjusted by both variables right?. So, it could also be linked to other scenarios: 1) genetic factors linked to high cholesterol levels (50% of levels are inherited), for example ACE2 gene polymorphisms have been correlated to both hypercholesterolemia and high ACE2 expression, which seems protective for SARS-2 infection; 2) direct effect of high cholesterol on the “lipid rafts”, which seem to play a role on SARS-2 infection mechanism and subsequent immune system activation

See some related manuscritps below

Lipid rafts play an important role in the early stage of severe acute respiratory syndrome-coronavirus life cycle

#8

I dont think there is likely a protective effect and one of the reasons is that if you look at the results you can see for e.g. sidiap covid that the “with outcome” mean is 0.2167 and the “no outcome” mean is 0.1105. So we see in the population stratified by outcome, those with the outcome have more hyperlipidemia than those without.
I would be surprised looking at this information that hyperlipidemia was actually protective but based on the research done here i wouldnt feel confident making a statement either way. Just that as part of a well evidenced and robust model it happens to have a negative coefficient.

#9

Hi: interesting debate. Just to understand: the “by outcome” analyses are univariate comparisons, correct? So this should correspond to the unadjusted regression coefficient for hyperlipidemia. The coefficient is reversed in the final model then?

Anyway the topic warrants some discussion. Better prediction models have causal factors in it we might claim; biology trumps local epiphenomena ?

BW, Ewout

t