In European Journal of Clinical Pharmacology

Online Resource for “Evaluation of methods to estimate missing days’ supply within pharmacy data of the Clinical Practice Research Datalink (CPRD) and The Health Improvement Network (THIN)”

in European Journal of Clinical Pharmacology

byKirsten J. Lum1,2, Craig W. Newcomb1, Jason A. Roy1,2, Dena M. Carbonari1,2, M. Elle Saine1,2, Serena Cardillo3, Harshvinder Bhullar4, Arlene M. Gallagher5, and Vincent Lo Re III*1,2,3

1Center for Clinical Epidemiology and Biostatistics, Department of Biostatistics and Epidemiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States

2Center for Pharmacoepidemiology Research and Training, Department of Biostatistics and Epidemiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States

3Department of Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania, United States

4IMS Health, London, United Kingdom

5Clinical Practice Research Datalink, London, United Kingdom

*Email for corresponding author:

Table 1 Mode days’ supply and percentage of prescriptions with known days’ supply equal to mode days’ supply by oral anti-diabetic drug and database.

CPRD
(7,486,695 OAD prescriptions) / THIN
(6,381,023 OAD prescriptions)
OAD class, medications / Mode days’ supply / % of OAD-specificprescriptions
with mode days’ supply / Mode days’ supply / % of OAD-specificprescriptions
with mode days’ supply
Overall / 28 / 59 / 28 / 56
Non Coformulated Tablets
85Alpha-glucosidase inhibitors
Acarbose / 28 / 35 / 30 / 39
Biguanides
Metformin / 28 / 60 / 28 / 58
Dipeptidyl peptidase IV inhibitors
Sitagliptin / 28 / 65 / 28 / 62
Saxagliptin / 28 / 68 / 28 / 69
Vildagliptin / 28 / 72 / 28 / 67
Linagliptin / 28 / 65 / 28 / 62
Metglitinides
Repaglinide / 28 / 37 / 28 / 33
Nateglinide / 28 / 70 / 28 / 59
Sodium-glucose cotransporter 2 inhibitors
Dapagliflozin / 28 / 71 / 28 / 66
Sulfonylureas 1st generation
Glimepiride / 30 / 38 / 30 / 34
Tolbutamide / 28 / 70 / 28 / 66
Sulfonylureas 2nd generation
Gliclazide / 28 / 55 / 28 / 54
Glipizide / 28 / 48 / 56 / 46
Glyburide / 28 / 59 / 28 / 57
Thiazolidinediones
Pioglitazone / 28 / 63 / 28 / 60
Rosiglitazone / 28 / 64 / 28 / 58
CPRD
(7,486,695 OAD prescriptions) / THIN
(6,381,023 OAD prescriptions)
Coformulated OAD Tablets / Mode days’ supply / % of OAD-specificprescriptions
with mode days’ supply / Mode days’ supply / % of OAD-specificprescriptions
with mode days’ supply
Metformin and Pioglitazone / 28 / 65 / 28 / 68
Metformin and Rosiglitazone / 28 / 67 / 28 / 66
Metformin and Vildagliptin / 30 / 58 / 30 / 56
Metformin and Sitagliptin / 28 / 67 / 28 / 64
Metformin and Linagliptin / 28 / 86 / 28 / 79
Metformin and Saxagliptin / 28 / 77 / 28 / 71

Abbreviations: CPRDClinical Practice Research Datalink,OAD oral anti-diabetic drug, THIN The Health Improvement Network

Table 2 Accuracy (calculated as: estimated days’ supply - known days’ supply) of three methods forestimation of days’ supply: 1) impute 28 days’ supply (Method 1), 2) impute mode number of tablets per day by drug strength and number of tablets per prescription (Method 2), and 3) impute number of tablets per day by drug strength, number of tablets per prescription, and several patient characteristics via a machine learning algorithma (Method 3), for oral anti-diabetic drug (OAD) medications not shown in Table 3 of manuscript. For each OAD, a random sample of 10,000 prescription records was drawn from prescriptions with known days’ supply separately within the Clinical Practice Research Datalinkand The Health Improvement Network(Oct 2009 – Dec 2013). Out of each sample, 75% of the prescriptions were used for training Methods 2 and 3 and 25% for testing Methods 2 and 3.

Most common
number of
tablets per dayb / OAD / Method / Accuracy: Estimated days’ supply – Known days’ supply
CPRD / THIN
Exactly
Correct
(%) / Within
10 days
(%) / Within
28 days
(%) / Exactly
Correct
(%) / Within
10 days
(%) / Within
28 days
(%)
1 / 69.7 / 69.9 / 99.1
1 / Dapagliflozinc / 2 / 99.8 / 99.8 / 100
3 / 99.8 / 99.8 / 100
1 / 22.4 / 60.9 / 79.8 / 25.7 / 61.1 / 78.3
1 / Glimepiride / 2 / 92.2 / 92.4 / 96.2 / 93.8 / 93.9 / 96.5
3 / 92.7 / 92.8 / 96.5 / 95.0 / 95.1 / 97.3
1 / 66.0 / 66.2 / 97.4 / 61.9 / 61.9 / 97.5
1 / Linagliptin / 2 / 99.6 / 99.9 / 100 / 99.2 / 99.9 / 100
3 / 99.8 / 99.9 / 100 / 100 / 100 / 100
1 / 62.7 / 62.8 / 97.7 / 60.1 / 60.3 / 97.7
1 / Pioglitazone / 2 / 99.2 / 99.2 / 99.9 / 99.4 / 99.4 / 100
3 / 99.2 / 99.2 / 99.9 / 99.4 / 99.4 / 100
1 / 64.4 / 64.9 / 97.3 / 58.5 / 59.2 / 96.7
1 / Rosiglitazone / 2 / 93.2 / 93.4 / 99.9 / 92.1 / 92.3 / 100
3 / 94.4 / 94.5 / 99.9 / 93.4 / 93.6 / 99.9
1 / 69.8 / 70.0 / 97.5 / 69.3 / 69.3 / 98.0
1 / Saxagliptin / 2 / 99.7 / 99.7 / 100 / 99.8 / 99.8 / 100
3 / 99.7 / 99.7 / 99.9 / 99.8 / 99.8 / 100
Most common
number of
tablets per dayb / OAD / Method / Accuracy: Estimated days’ supply – Known days’ supply
CPRD / THIN
Exactly
Correct
(%) / Within
10 days
(%) / Within
28 days
(%) / Exactly
Correct
(%) / Within
10 days
(%) / Within
28 days
(%)
1
2 / Metformin andlinagliptin
coformulatedtabletsd / 2
3
1 / 64.3 / 64.8 / 97.3 / 68.1 / 68.8 / 97.0
2 / Metformin andpioglitazone
coformulated tablets / 2 / 97.3 / 97.5 / 99.9 / 97.8 / 98.2 / 100
3 / 97.5 / 97.8 / 99.9 / 98.4 / 98.5 / 100
1 / 66.3 / 66.9 / 97.2 / 66.4 / 67.3 / 98.2
2 / Metformin androsiglitazone
coformulated tablets / 2 / 95.2 / 95.5 / 99.6 / 94.5 / 94.7 / 99.8
3 / 96.7 / 97 / 99.9 / 96.8 / 96.9 / 100
1
2 / Metformin andsaxagliptin
coformulatedtabletse / 2
3
1 / 13.5 / 70.2 / 81.2 / 15 / 70.9 / 83.4
2 / Metformin andvildagliptin
coformulated tablets / 2 / 99.2 / 99.2 / 99.7 / 99.6 / 99.6 / 99.9
3 / 99.6 / 99.6 / 99.7 / 99.7 / 99.7 / 99.9
3 / Acarbose / 1 / 34.9 / 68.7 / 87.9 / 29.4 / 68.4 / 88.3
2 / 84.6 / 85.9 / 93.2 / 82.9 / 85.6 / 92.9
3 / 96.8 / 96.9 / 98.7 / 97.2 / 97.3 / 98.9
1 / 69.6 / 71.1 / 96.6 / 59.4 / 62.3 / 95.7
3 / Nateglinide / 2 / 92.8 / 93.1 / 98.9 / 92.1 / 92.1 / 98.8
3 / 97.7 / 97.7 / 99.8 / 97.8 / 97.9 / 99.6
Most common
number of
tablets per dayb / OAD / Method / Accuracy: Estimated days’ supply – Known days’ supply
CPRD / THIN
Exactly
Correct
(%) / Within
10 days
(%) / Within
28 days
(%) / Exactly
Correct
(%) / Within
10 days
(%) / Within
28 days
(%)
1 / 37.4 / 70.8 / 90.0 / 31.9 / 64.9 / 87.8
3 / Repaglinide / 2 / 81.9 / 86.2 / 94.4 / 79.3 / 81.8 / 91.4
3 / 96.1 / 96.3 / 98.4 / 95.9 / 96.1 / 98.0
1 / 53.2 / 58.8 / 94.5 / 54.5 / 59.9 / 94.1
Variable / Gliclazide / 2 / 70.9 / 72.2 / 95.8 / 71.0 / 71.7 / 95.4
3 / 66.5 / 69.7 / 95.9 / 65.3 / 67.8 / 95.5
1 / 48.0 / 49.6 / 97 / 45.7 / 47.3 / 97.9
Variable / Glipizide / 2 / 63.6 / 65.2 / 98.3 / 63.7 / 64.6 / 99.1
3 / 79.7 / 80.5 / 98.7 / 78.4 / 79.0 / 99.0
1 / 69.4 / 70.9 / 97.5 / 66.9 / 68.1 / 96.9
Variable / Tolbutamidef / 2 / 78.7 / 79.8 / 99.3 / 74.6 / 75.4 / 98.7
3 / 94.6 / 95.2 / 99.7 / 94.8 / 95.1 / 99.6

Abbreviations: CPRD Clinical Practice Research Datalink, THINThe Health Improvement Network

aVariables in training model: drug strength, number of tablets per prescription, patient age, indicator of co-therapy within prior 90 days, and diabetic complications. For coformulated tablets, indicator of co-therapy within prior 90 days was not included in training model because it all prescriptions for coformulated tablets were categorized as co-therapy.

bMost common number of tablets per day is determined by largest percentage of prescriptions greater than 50% given in Table 2.

cMethods could not be compared in CPRD because all prescriptions in training data were prescribed 1 tablet per day.

dMethods could not be compared in CPRD or THIN because all prescriptions in training data were prescribed 2 tablets per day.

eMethods could not be compared in CPRD or THIN because all prescriptions in training data were prescribed 2 tablets per day.

fDrug strength was not used in Methods 2 and 3 because all prescriptions for tolbutamide in the training data were for the same strength

Table 3Impact of choice of method for estimating days’ supply on exposure time, total number of events, and incidence rate per 1,000 years for study samples from the Clinical Practice Research Datalink (panel a) and The Health Improvement Network(panel b; Oct 2009 – Dec 2013). Three methods were compared for estimating days’ supply when missing: 1) impute 28 days’ supply, 2) impute mode number of tablets per day by drug strength and number of tablets per prescription, and 3) impute number of tablets per day by drug strength, number of tablets per prescription, and several patient characteristics via a machine learning algorithma. Analyses were performed separately for each medication.

a) CPRD (n=201,740 OAD users)
Exposure time
(in person years) / Total
no. of events / IR per 1,000 years
(95% CI)
Method / Method / Method
Most common
number of
tablets per dayb / OAD / %rx
missing
number of
tablets per day / Total
no. of
users / %
of
users / 1 / 2 / 3 / 1 / 2 / 3 / 1 / 2 / 3
1 / Sitagliptin / 9 / 25,723 / 13 / 35,064 / 35,703 / 35,703 / 160 / 164 / 164 / 4.56
(3.88-5.33) / 4.59
(3.92-5.35) / 4.59
(3.92-5.35)
1 / Pioglitazone / 8 / 23,883 / 12 / 45,166 / 45,991 / 45,986 / 171 / 175 / 175 / 3.79
(3.24-4.40) / 3.81
(3.26-4.41) / 3.81
(3.26-4.41)
1 / Glimepiride / 13 / 7,972 / 4 / 15,647 / 16,075 / 16,092 / 102 / 105 / 105 / 6.52
(5.32-7.91) / 6.53
(5.34-7.91) / 6.52
(5.34-7.90)
1 / Saxagliptin / 9 / 4,482 / 2 / 4,573 / 4,631 / 4,631 / 24 / 24 / 24 / 5.25
(3.36-7.81) / 5.18
(3.32-7.71) / 5.18
(3.32-7.71)
1 / Rosiglitazone / 8 / 7,159 / 4 / 5,122 / 5,230 / 5,233 / 12 / 12 / 12 / 2.34
(1.21-4.09) / 2.29
(1.19-4.01) / 2.29
(1.19-4.01)
1 / Linagliptin / 10 / 2,551 / 1 / 1,577 / 1,594 / 1,594 / 15 / 16 / 16 / 9.51
(5.32-15.69) / 10.04
(5.74-16.30) / 10.04
(5.74-16.30)
1 / Dapagliflozinc / 3 / 200 / <0.5 / 54 / 54 / 54 / 1 / 1 / 1 / 18.62
(0.47-103.72) / 18.62
(0.47-103.72) / 18.62
(0.47-103.72)
CPRD (n=201,740 OAD users) (continued)
Exposure time
(in person years) / Total
no. of events / IR per 1,000 years
(95% CI)
Method / Method / Method
Most common
number of
tablets per dayb / OAD / %rx
missing
number of
tablets per day / Total
no. of
users / %
of
users / 1 / 2 / 3 / 1 / 2 / 3 / 1 / 2 / 3
2 / Vildagliptin / 14 / 1,885 / 1 / 2,713 / 2,777 / 2,778 / 9 / 9 / 9 / 3.32
(1.52-6.30) / 3.24
(1.48-6.15) / 3.24
(1.48-6.15)
3 / Acarbose / 21 / 962 / <0.5 / 1,477 / 1,525 / 1,524 / 11 / 12 / 12 / 7.45
(3.72-13.33) / 7.87
(4.07-13.75) / 7.87
(4.07-13.75)
3 / Nateglinide / 26 / 270 / <0.5 / 416 / 435 / 430 / 0 / 0 / 0 / 0.00
(0.00-7.19) / 0.00
(0.00-6.88) / 0.00
(0.00-6.96)
3 / Repaglinide / 51 / 1,180 / 1 / 1,561 / 1,665 / 1,651 / 9 / 9 / 8 / 5.77
(2.64-10.94) / 5.41
(2.47-10.26) / 4.85
(2.09-9.55)
Variable / Metformin / 24 / 181,262 / 90 / 373,604 / 388,283 / 389,798 / 1,696 / 1,757 / 1,765 / 4.54
(4.33-4.76) / 4.53
(4.32-4.74) / 4.53
(4.32-4.74)
Variable / Gliclazide / 31 / 78,038 / 39 / 142,289 / 145,945 / 147,517 / 869 / 890 / 895 / 6.11
(5.71-6.53) / 6.10
(5.70-6.51) / 6.07
(5.68-6.48)
Variable / Glipizide / 32 / 2,839 / 1 / 5,298 / 5,548 / 5,749 / 41 / 42 / 43 / 7.74
(5.55-10.50) / 7.57
(5.46-10.23) / 7.48
(5.41-10.07)
Variable / Glyburide / 19 / 1,367 / 1 / 2,411 / 2,530 / 2,514 / 13 / 14 / 13 / 5.39
(2.87-9.22) / 5.53
(3.03-9.28) / 5.17
(2.75-8.84)
Variable / Tolbutamide / 27 / 853 / <0.5 / 1,529 / 1,552 / 1,577 / 13 / 13 / 13 / 8.50
(4.53-14.54) / 8.38
(4.46-14.33) / 8.24
(4.39-14.09)

Abbreviations: CI confidence interval, CPRDClinical Practice Research Datalink,IR incidence rate, OAD oral anti-diabetic drug, rxprescription

aVariables in training model: drug strength, number of tablets per prescription, patient age, indicator of co-therapywithin prior 90 days, and diabetic complications.

bMost common number of tablets per day is determined by largest percentage of prescriptions greater than 50% given in Table 2.

cFordapagliflozin, incidence rates using Methods 2 and 3 are identical because all prescriptions in training data were prescribed 1 tablet per day.

b) THIN (n=172,163 OAD users)
Exposure time
(in person years) / Total
no. of events / IR per 1,000 years
(95% CI)
Method / Method / Method
Most common
number of
tablets per dayb / OAD / %rx
missing
number of
tablets per day / Total
no. of
users / %
of
users / 1 / 2 / 3 / 1 / 2 / 3 / 1 / 2 / 3
1 / Sitagliptin / 14 / 22,231 / 13 / 29,827 / 30,634 / 30,635 / 149 / 153 / 153 / 5.00
(4.23-5.87) / 4.99
(4.23-5.85) / 4.99
(4.23-5.85)
1 / Pioglitazone / 13 / 20,234 / 12 / 38,213 / 39,169 / 39,166 / 136 / 139 / 139 / 3.56
(2.99-4.21) / 3.55
(2.98-4.19) / 3.55
(2.98-4.19)
1 / Glimepiride / 22 / 6,681 / 4 / 12,862 / 13,493 / 13,532 / 80 / 86 / 86 / 6.22
(4.93-7.74) / 6.37
(5.10-7.87) / 6.36
(5.08-7.85)
1 / Saxagliptin / 13 / 3,771 / 2 / 3,908 / 3,979 / 3,979 / 16 / 16 / 16 / 4.09
(2.34-6.65) / 4.02
(2.30-6.53) / 4.02
(2.30-6.53)
1 / Rosiglitazone / 16 / 6,312 / 4 / 4,445 / 4,613 / 4,614 / 11 / 11 / 11 / 2.47
(1.24-4.43) / 2.38
(1.19-4.27) / 2.38
(1.19-4.27)
1 / Linagliptin / 19 / 2,341 / 1 / 1,398 / 1,431 / 1,431 / 10 / 11 / 11 / 7.15
(3.43-13.15) / 7.69
(3.84-13.75) / 7.69
(3.84-13.75)
1 / Dapagliflozin / 10 / 717 / <0.5 / 217 / 220 / 220 / 1 / 1 / 1 / 4.61
(0.12-25.67) / 4.55
(0.12-25.36) / 4.55
(0.12-25.36)
2 / Vildagliptin / 14 / 2,053 / 1 / 3,159 / 3,213 / 3,224 / 9 / 9 / 9 / 2.85
(1.30-5.41) / 2.80
(1.28-5.32) / 2.79
(1.28-5.30)
3 / Acarbose / 27 / 791 / <0.5 / 1,165 / 1,207 / 1,202 / 4 / 5 / 5 / 3.43
(0.94-8.79) / 4.14
(1.34-9.66) / 4.16
(1.35-9.71)
3 / Nateglinide / 29 / 223 / <0.5 / 355 / 369 / 366 / 1 / 1 / 1 / 2.82
(0.07-15.71) / 2.71
(0.07-15.10) / 2.73
(0.07-15.21)
3 / Repaglinide / 58 / 915 / 1 / 1,261 / 1,350 / 1,347 / 5 / 5 / 5 / 3.97
(1.29-9.25) / 3.70
(1.20-8.64) / 3.71
(1.21-8.66)
THIN (n=172,163 OAD users) (continued)
Exposure time
(in person years) / Total
no. of events / IR per 1,000 years
(95% CI)
Method / Method / Method
Most common
number of
tablets per dayb / OAD / %rx
missing
number of
tablets per day / Total
no. of
users / %
of
users / 1 / 2 / 3 / 1 / 2 / 3 / 1 / 2 / 3
Variable / Metformin / 34 / 154,389 / 90 / 314,501 / 331,491 / 332,728 / 1,380 / 1,456 / 1,462 / 4.39
(4.16-4.63) / 4.39
(4.17-4.62) / 4.39
(4.17-4.63)
Variable / Gliclazide / 40 / 66,320 / 39 / 119,533 / 123,879 / 124,161 / 704 / 723 / 719 / 5.89
(5.46-6.34) / 5.84
(5.42-6.28) / 5.79
(5.38-6.23)
Variable / Glipizide / 42 / 2,687 / 2 / 5,035 / 5,428 / 5,564 / 35 / 35 / 37 / 6.95
(4.84-9.67) / 6.45
(4.49-8.97) / 6.65
(4.68-9.17)
Variable / Glyburide / 26 / 1,238 / 1 / 2,153 / 2,261 / 2,304 / 12 / 13 / 12 / 5.57
(2.88-9.74) / 5.75
(3.06-9.83) / 5.21
(2.69-9.10)
Variable / Tolbutamide / 41 / 758 / <0.5 / 1,391 / 1,417 / 1,457 / 14 / 14 / 14 / 10.07
(5.50-16.89) / 9.88
(5.40-16.57) / 9.61
(5.25-16.12)

Abbreviations:CI confidence interval, IR incidence rate, OAD oral anti-diabetic drug, rxprescription, THIN The Health Improvement Network

aVariables in training model: drug strength, number of tablets per prescription, patient age, indicator of co-therapywithin prior 90 days, and diabetic complications.

bMost common number of tablets per day is determined by largest percentage of prescriptions greater than 50% given in Table 2.

Table 4 Results of test for difference in accuracy comparing Methods 2 and 3 for estimation of days’ supply, where Method 2 is to impute mode number of tablets per day by drug strength and number of tablets per prescription andMethod 3 is to impute number of tablets per day by drug strength, number of tablets per prescription, and several patient characteristics via a machine learning algorithma. For each OAD, a random sample of 10,000 prescription records was drawn from prescriptions with known days’ supply separately within the Clinical Practice Research Datalink and The Health Improvement Network(Oct 2009 – Dec 2013). Out of each sample, 75% of the prescriptions were used for training Methods 2 and 3 and 25% for testing Methods 2 and 3.

CPRD / THIN
Most
common
number
of
tablets
per dayb / OAD / Discordant pairs / McNemar’s
exact
test
p-valuec / Discordant pairs / McNemar’s
exact
test
p-valuec
Cell b / Cell c / Cell b / Cell c
1 / Dapagliflozind / 1 / 1 / 1.00
1 / Glimepiride / 48 / 61 / 0.25 / 52 / 82 / 0.01
1 / Linagliptin / 0 / 7 / 0.02 / 0 / 19 / <0.01
1 / Pioglitazone / 0 / 0 / 1.00 / 1 / 2 / 1.00
1 / Rosiglitazone / 12 / 43 / <0.01 / 38 / 71 / <0.01
1 / Saxagliptin / 1 / 0 / 1.00 / 0 / 0 / 1.00
1 / Sitagliptin / 0 / 0 / 1.00 / 1 / 1 / 1.00
2 / Vildagliptin / 21 / 61 / <0.01 / 32 / 108 / <0.01
2 / Metformin-linagliptin
coformulatedtabletse
2 / Metformin-pioglitazone
coformulated tablets / 10 / 16 / 0.33 / 1 / 17 / <0.01
2 / Metformin-rosiglitazonecoformulated tablets / 16 / 52 / <0.01 / 17 / 75 / <0.01
2 / Metformin- saxagliptincoformulatedtabletsf
2 / Metformin- sitagliptincoformulated tablets / 1 / 3 / 0.63 / 0 / 13 / <0.01
2 / Metformin- vildagliptincoformulated tablets / 0 / 10 / <0.01 / 1 / 3 / 0.63
3 / Acarbose / 11 / 313 / <0.01 / 15 / 370 / <0.01
3 / Nateglinide / 6 / 54 / <0.01 / 8 / 51 / <0.01
3 / Repaglinide / 29 / 381 / <0.01 / 32 / 323 / <0.01
Variable / Gliclazide / 247 / 136 / <0.01 / 152 / 12 / <0.01
Variable / Glipizide / 172 / 572 / <0.01 / 224 / 590 / <0.01
Variable / Glyburide / 88 / 484 / <0.01 / 61 / 469 / <0.01
Variable / Metformin / 226 / 176 / 0.014 / 196 / 231 / 0.10
Variable / Tolbutamide / 38 / 433 / <0.01 / 41 / 544 / <0.01

Abbreviations: CPRD Clinical Practice Research Datalink, OAD oral anti-diabetic drug, THIN The Health Improvement Network

Cell b is the number of prescriptions in the test data for which days’ supply was exactly correct by Method 2 but incorrect by Method 3.

Cell c is the number of prescriptions in the test data for which days’ supply was exactly correct by Method 3 but incorrect by Method 2.

bMost common number of tablets per day is determined by largest percentage of prescriptions greater than 50% given in Table 2.

cNull hypothesis was no difference in accuracy between Methods 2 and 3. Separate tests were conducted for each OAD and database.

dMethods could not be compared in CPRD because all prescriptions in training data were prescribed 1 tablet per day.

eMethods could not be compared in CPRD or THIN because all prescriptions in training data were prescribed 2 tablets per day.

fMethods could not be compared in CPRD or THIN because all prescriptions in training data were prescribed 2 tablets per day.