Table 1 Articles that mention the most usual biases described in observational studies of pharmacoepidemiologic databases

Confounding

The measure of association between treatment and outcome is distorted by the effect of one or more variables, which are also risk factors for the outcome of interest

[1,2,3, 6, 14,15,16, 18, 22, 40, 41, 57, 58, 62, 80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139]

63.2

 Confounding by indication^a

The clinical condition that determined the prescription of the treatment is associated with the effect, acting as a confounding factor (e.g. a worse disease status at baseline: confounding by disease severity)

[3, 6, 18, 22, 40, 41, 57, 80, 82, 84, 86, 87, 89, 90, 92, 96, 97, 99, 100, 104, 106, 107, 110, 111, 113, 114, 116, 118, 120, 122, 126, 128,129,130,131, 133, 134, 138]

32.5

 Time-dependent confounding

A variable that can vary with time acts as a confounding factor between the current exposure and outcome, and as an intermediary between prior and current exposure

[40, 41, 57, 58, 81, 92, 104]

6.0

 Unmeasured/residual confounding

There is not enough information about all the relevant confounding factors known, unknown or difficult to measure (e.g. frailty). If confounding cannot be completely controlled for, the residual confounding effect of some factors remains in the final effect that is observed

[1,2,3, 6, 14, 15, 18, 58, 62, 80,81,82,83, 86, 89, 91,92,93, 96, 101, 103, 108, 110, 113, 116, 119, 125, 127, 130, 132, 134, 136, 139]

28.2

  Healthy user/adherer effect

Access to health care resources is associated with a higher level of education and health-seeking behavior. Furthermore, patients who comply with the treatment during prolonged periods of time tend to be healthier

[2, 18, 91, 96, 125, 127]

5.1

Selection bias

The study sample population is not representative of the target population to which the results will be extrapolated

[2, 16, 18, 22, 40, 41, 54, 57, 58, 63, 81, 83, 84, 87, 88, 90, 91, 93,94,95, 99, 101,102,103, 105, 107,108,109, 111,112,113, 115,116,117,118,119, 121, 122, 124, 125, 135,136,137, 140,141,142,143,144,145,146,147,148,149,150,151]

47.0

 Protopathic bias

The treatment is associated with subclinical disease stages (an early manifestation of the still undiagnosed condition under study gives rise to prescription of the treatment)

[40, 41, 81, 109]

3.4

 Losses to follow-up (informative censoring)

The mechanism that triggers discontinuity of the treatment is associated with the risk of observing the outcome of interest

[40, 41, 116]

2.6

 Depletion of susceptibles (prevalent user bias)

The inclusion of prevalent instead of incident users entails insufficient verification of the adverse effects that occur at the beginning of treatment (those susceptible to the adverse effect have interrupted the treatment)

[2, 40, 41, 57, 83, 90, 99, 107, 111, 116, 118, 148]

10.3

 Missing data

In multivariate analyses, such as regression models, observations that lack one or more of the values of a variable included in the model tend to be eliminated

[58, 63, 87, 93, 94, 108, 112, 116, 119, 125, 135,136,137, 140, 141, 143,144,145,146,147, 151]

17.9

Measurement bias

Data on true exposures, outcomes and other variables are recorded in the form of indicators (observed measures) that do not accurately reflect reality

[2, 3, 6, 7, 16, 40, 41, 54, 55, 58, 87, 88, 91, 93, 94, 96, 101, 105, 108, 110, 112, 114, 115, 117, 119, 121, 124, 125, 130, 135,136,137,138, 140, 141, 143, 144, 146, 147, 149, 151,152,153,154,155,156,157,158,159,160,161,162,163,164]

46.2

 Misclassification bias

The association between treatment and outcome is distorted by systematic errors, due to the way in which the variables of interest are measured in comparison groups

[2, 3, 6, 7, 16, 40, 41, 54, 55, 58, 87, 88, 91, 93, 94, 96, 101, 105, 108, 110, 112, 114, 115, 119, 121, 125, 130, 135,136,137,138, 140, 141, 143, 144, 146, 147, 149, 152,153,154,155,156,157,158,159,160,161,162,163,164]

43.6

  Misclassification of exposure

The measure of exposure of a given treatment is not an exact reflection of its real use (e.g. flawed measurement, non-compliance with treatment, inappropriate use of time windows)

[2, 3, 16, 40, 41, 54, 55, 58, 87, 91, 93, 94, 96, 101, 110, 119, 121, 130, 138, 140, 146, 147, 152, 154, 156, 158, 159, 164]

23.9

  Misclassification of outcome

Error in the diagnosis (e.g. clinical ambiguity, non-uniform coding)

[2, 3, 6, 7, 16, 40, 41, 54, 58, 87, 91, 93, 94, 96, 101, 110, 112, 114, 121, 125, 135,136,137, 141, 143, 149, 153, 155, 157, 160,161,162,163]

28.2

Time-related bias

Follow-up time and exposure status are inadequately taken into account in the study-design or analysis stages

[2, 7, 40, 41, 57, 68,69,70,71,72,73,74,75, 77, 83, 86, 87, 90, 99, 101, 105,106,107, 111, 114, 118, 128, 129, 133, 142, 165,166,167,168,169,170]

30.8

 Immortal time bias

A period of time (immortal) during which the study event cannot occur is included in the follow-up or is excluded from analysis due to an incorrect definition of the start of follow-up

[2, 7, 40, 41, 57, 68,69,70,71,72,73,74,75, 77, 83, 86, 87, 90, 99, 101, 106, 107, 111, 114, 118, 128, 129, 133, 166, 167]

25.6

 Immeasurable time bias

A period of time (immeasurable) during follow-up is ignored and thus misclassified as unexposed period, since outpatient prescriptions that define exposure cannot occur (e.g. serious chronic diseases that require extensive use of medications and multiple hospitalizations)

[142, 165, 168, 170]

3.4

 Time-window bias

The use of time-windows of different lengths between cases and controls to define time-dependent exposures prevents subjects from having the same opportunity time to receive prescriptions

[90, 106, 169]

2.6

 Time-lag bias

Comparisons are conducted of treatments given at different stages of the disease, which inherently introduces bias related to disease duration and progression

[106]

0.9