Bastiaan Sallevelt

126 CHAPTER 2.3 We followed the original Irish STOPP/START criteria as closely as possible. By providing the actual algorithms and code dictionary with this publication, users are given the resources to make different choices about included ICD, ICPC and ATC codes or change cut-off levels for laboratory measurements following local guidelines. Therefore, these algorithms could serve as a template for applying STOPP/START criteria version 2 (or a subset of the criteria) to any software application. Limitations Despite maximal effort to be as complete and punctual as possible, several limitations to this study need to be addressed. For the algorithms presented here, the original Irish STOPP/START criteria, as published in Age & Aging in 2015, were used [13]. However, many local versions of these criteria exist in different countries based on variations in local guidelines. This may reduce the applicability of the algorithms to the country-specific situation. However, by providing our algorithms accompanied by a code dictionary including all the mentioned and coded diseases and medications per criterion, users can easily adapt the algorithm to match their local versions of STOPP/START. In our coding strategy, we decided to translate the criteria as accurately as conceivable, assuming that data registration in research databases and patients’ health records is carried out perfectly by health care professionals. For instance, if a criterion is restricted to the condition of ‘chronic atrial fibrillation’, as is the case in START A1 and A2, we have coded this as the exact matching term ICD-10 I48.2: ‘chronic atrial fibrillation’ instead of I48: ‘atrial fibrillation and flutter’. When applying the algorithm to a database using ICD-10 codes, this decision may lead to under detection of START A1 and A2, as atrial fibrillation is not always documented as either chronic or paroxysmal. Physicians and other health care professionals (HCP) should be encouraged to accurately code diseases and diagnoses according to international classification databases to enable data extraction. Educational programs to train HCPs in meticulous registration is crucial to successfully implement coded algorithms into electronic health records. Furthermore, expert based choices had to be made in cases where criteria were ambiguous or not matching the database terminology. For instance, opioids are not classified as either high or low-potency (START H1) in the WHO-ATC database and required expert consensus. In addition, cut-off values needed to be determined where these were not explicitly mentioned in the criteria. The potential hazard of hyperkalemia is addressed in several criteria, like STOPP B11: ‘ACE inhibitor or Angiotensin Receptor Blockers in patients with hyperkalemia’. We defined hyperkalemia as ≥ 5.0mmol/L, a generally accepted cut-off value within laboratory