The case-crossover design via penalized regression

Table 1 Number of drugs correctly identified under new reclassification rule by each method

Original Classification Rule for new reclassification	Highly suspected (n = 9) log OR > 2.4	Suspected (n = 10) 0.45 < log OR < 2.4	Other (n = 11) log OR < 0.45	Total (n = 30)
Univariable CLR	5	7	7	19
Multivariable CLR	6	2	9	17
Lasso	5	8	9	22
Elastic net	4	8	9	21
Bolasso	4	6	9	19
Sublasso	4	8	9	21
Random lasso	6	6	10	22
Firth correction	5	8	7	20
Case-control^a	9	10	8	27

^aNote that even though the case–control study was the basis for the risk classification in Mockenhaupt et al. [7], its estimates do not fully agree with the reclassification scheme used by us. This is because experts classified some of the drugs differently than its risk estimates would suggest based on their experience and opinion, whereas we employ a reclassification scheme using only the estimates

ISSN: 1471-2288