Thread by @deeksj

Sorry - but there is a dreadful mistake made in computing the sensitivity and specificty of LFT in this report. If you look at Figure 32 (day 1 for example) the estimates of sens and spec are based on a subsample of the study with 364 LFT+ve and 686 LFT -ve. 34.7% are LFT+VE

https://twitter.com/CwmTafMorgannwg/status/1373961796080562186

However, in the whole sample 33,315 LFD tests were completed across 12 centres. Of these, 763 were positive, representing a positivity rate of 2.3%.

Thus the sample used in the test accuracy study is biased to include many more LFT+ves (34.7% compared to 2.3%).

This means overcounting of true positives and false positives, undercounting of true negatives and false negatives - but a large order of magnitude.

Sensitivity and specificity estimates are badly affected

You can work out the sampling weights.

Of the LFT positives 364 out of 763 were included (47.7%).

Of the LFT negatives 686 out of 32552 were included (2.1%).

Sensitivity is computed as TP/(TP+FN). Now we need to reweight TP and FN by inverse of sampling probabilities.

Quoted figure was 324 / (324+67) =82.9%

(TP/0.477) / [ (TP/0.477)+(FN/0.021) ]

(324/0.477) / [ (324/0.477) + (67/0.021) ]

679 / [679+3190] = 17.5%

MASSIVE ERROR

Specificity is computed as TN/(TN+FP). Now we need to reweight TN and FP by inverse of sampling probabilities.

Quoted figure was 619 / (619+40) =93.9%

(TN/0.021) / [ (FP/0.477)+(TN/0.021) ]

(619/0.021) / [ (40/0.477) + (619/0.021) ]

29476 / [29476+84] = 99.7%

MASSIVE ERROR

Also can you please explain how you recruited LFT-ves for PCR testing ? Your report states LFT+ves were given a home test to do, but no mention of who underwent PCR testing in the LFT-ves. We need to know who and how to be able to judge if represetative.

These errors are so large I would suggest that this report is withdrawn and properly peer reviewed.

It is SERIOUSLY MISLEADING