However, its specificity value of 55%, was lower than GerontoNet’s at 65%. It must be noted however, that GerontoNet only considered cases of definite and probable ADRs for inclusion in their study, whereas possible ADRs were included in the present study. Other studies which have attempted to develop predictive models for adverse drug events have not been externally validated which restricts their applicability to other populations. Much work has been conducted over recent years to identify clinical risks associated with ADR. Interestingly, a number of these risks were not found to be significant predictors of ADR in this current study. For example, well known factors such as age, previous history of ADR, gender, heart failure, prior bleeding on admission, renal impairment, the use of certain drug classes, and abnormalities in certain laboratory parameters were not retained in our final model. The omission of these previously identified risk factors could possibly be explained by the inclusion criteria, which recruited a broader range of ages. Of the variables included in our final model, the number of medications prescribed during inpatient stay was the only variable that has been consistently identified as an ADR risk in all other validated studies. One variable previously identified by O’Connor et al to predict ADR risk is that of potentially inappropriate medicines. These were not included in this study as this variable was identified after our study had been conducted. This is the first study to include total length of stay, hyperlipidaemia and white cell count as ADR risk factors. Length of stay may be a proxy measure for the severity of the patients’ underlying illness, perhaps reflecting an increase in the number of prescribed medicines. However, the fact that the number of medications prescribed is also included as a risk factor in the BADRI model, and that, when tested, no association between the two variables was found, suggests that they (+)-JQ1 msds represent different mechanisms. Alternatively, an increased length of stay may reflect a deterioration in the patients clinical state, and a change in the pharmacokinetic or pharmacodynamic profile of the medication, leading to altered drug levels or response. It is also conceivable that a prolonged hospitalisation could increase the likelihood of experiencing, and detecting, an ADR, i.e. a posteriori observation, which questions the value of this variable as a useful predictor. We also found that a diagnosis of hyperlipidaemia significantly increased the likelihood of developing an ADR. The reason for its identification as a risk factor has yet to be determined, but may simply reflect the fact that individuals with a history of an abnormal lipid profile are at higher risk of cardiovascular disease, and as a consequence, may be receiving multiple drug therapies. An association between a high white blood cell count on admission and ADR risk was another novel risk factor identified in the current study. This may reflect that the patient is suffering from an infection or inflammatory condition which requires the subsequent use of potential harmful antibiotics or cardiac medications. Regardless of the cause of raised WCC, it is nonetheless a marker for increased susceptibility to ADR, and will alert the clinician to the patients increase risk of developing an adverse reaction to a medicine. Although there are several advantages of the BADRI score in predicting ADR risk, there are of course some limitations that need to be addressed. Causality assessment for ADR in the testing dataset was conducted using Naranjo’s algorithm, however, during the initial development of the model, the Hallas criteria was utilised to determine causality.