Bastiaan Sallevelt

117 Conversion of STOPP/START version 2 into coded algorithms A physician (CH; geriatric resident and PhD researcher) and a pharmacist (BS; hospital pharmacist in training and PhD researcher) were consulted as a validation panel, based on their experience with developing and implementing STOPP/START algorithms in a CDSS. During the third round, the validation and preparation panel (DdG, MM, CH and BS) reviewed and discussed all coded algorithms in three faceto-face meetings according to the coding principles, focusing both on content (i.e. completeness and consistency of incorporated ICD, ICPC and ATC codes) and on logic (i.e. the interrelationship of different items within one algorithm). Fourth round The validation panel applied the input of the experts to the algorithm and performed a functionality check for each criterion on logic, integrality and inter- and intra-item consistency using the defined coding principles. The draft version of the algorithm and the dictionary were updated accordingly. After consensus was reached regarding the content of the coded criteria, the ICD, ATC and LOINC based algorithms were implemented in a stand-alone, web-based CDSS (STRIP Assistant) [20]. This round was an ultimate test to verify whether the content and logic, as theoretically approved in the third round, would reveal any unexpected errors if used in a computer system. Therefore, all coded criteria were systematically tested in order to find false positive and false negative triggers, as well as logical errors within the algorithm. The conditions required to trigger an individual STOPP/START criterion were entered in the CDSS. If a specific criterion was not triggered while expected based on the data input into the CDSS, the algorithms were checked again to assess whether this was due to a coding problem based on content (i.e. ICD or ATCmismatch) or a logical problemwithin the algorithm itself. This process was repeated for all coded algorithms independently. A schematic representation of the approach is displayed in Figure 2. During this functionality check, it was found that the omission of exceptions within certain criteria generated false positive triggers if the algorithms were applied without any clinical judgement. For those criteria, the validation panel decided - in accordance with the experts - to add ‘optional (excluding) conditions’ to the algorithm, that were not actually present in the original STOPP/START criteria, to enhance (clinical) applicability of the algorithms. The adjusted set of algorithms was sent to all members of the expert panel for final approval. 2