Triggers of potential safety risks were reported in the anaesthesia literature 20 years ago.1 Trigger tools are sets of easily identified flags, occurrences or prompts that alert reviewers to situations where harm is thought to be more likely than in routine care.2Where there are electronic health records, applying both prospective and retrospective computer search algorithms for various triggers has been proposed as a method of identifying error and adverse events, especially in hospitals.3 Such searches provide a reasonably unbiased, systematic method of reviewing patient records to alert doctors and nurses to potentially risky situations and to provide measures of safety improvement as harm avoidance measures are implemented.The usefulness of identifying harm is that processes and systems within practices that may lead or contribute to harm, can be analysed and changed, if we knew what they were. To be effective in this role, triggers should be sensitive (i.e. identify all occasions of the trigger event occurring) and specific (i.e. not identify situations that seldom result in harm to patients). There are some reports of proposed triggers having sensitivity and specificity problems.4 This makes their use inefficient as on each occasion a trigger occurs, a manual review must be done to assess whether harm has occurred, and (if it has) its type and severity.If the potential for harm associated with a trigger is seldom realised and the trigger identifies a common situation, the labour associated with reviewing triggered cases may be a cost that overwhelms possible benefits. Reports of trigger tools being tested in UK primary care practices show that it is possible to review up to 20 records in a 2-3 hour session, and that 8-12 triggers may provide optimal balance between sensitivity, specificity, and feasibility for using as a routine safety improvement tool.5-7Despite reports of the development of primary care trigger tools, little is yet known about the practicalities of using them in practice and in New Zealand there are no reports of their uptake. We could find no research showing the role of trigger tools in documenting the underlying harm arising from care provided in general practice settings. As a result it has been difficult to extrapolate these trigger tools to our clinical context, understand the proportion of harm that might be identified if we used one of the existing trigger tools, and inform our decisions about making our primary care safer for patients.Because of the potential importance of triggers in protecting patient safety, we decided to test their use in a large general practice (>12,000 enrolled patients) situated in provincial New Zealand. The practices patients are mainly New Zealand European but Mori comprise 18% of its enrolled population. Its catchment includes both urban and rural areas.We aimed to establish what trigger tool worked for us, which triggers were most useful, and whether we could derive a process that would be practical for us to use routinely.Methods Possible triggers were identified from reviewing the literature of triggers tested in primary care and a focus group of two general practitioners, two pharmacists and one practice nurse decided on the 36 triggers for initial use (Table 1). The focus group was facilitated by the local Primary Healthcare Organisations (PHOs) quality improvement leader. In New Zealand, PHOs are responsible for the funding, quality improvement and clinical governance of primary care. We calculated that we needed to review the records of 170 patients, based on an assumption that the background harm rate in primary care is 5% and with 90% power to detect harm. To be included in the review, patients had to be registered with the practice for \u226512 months and have at least one visit with a general practitioner in 2011. We decided to include all ages in the cohort (other studies of primary care trigger tools had excluded children) and that 50% of reviews would be of Mori patients records. Records were reviewed from patients randomly selected from the practices January 2011 patient register. The trigger tool was applied by two teams of reviewers. One team consisted of a general practitioner and a community pharmacist and the other team was a general practitioner and a practice nurse. The teams separately reviewed each patient record for the presence of a trigger. If one was present, indication of harm relating to that trigger was then sought. Table 1. The initial trigger tool and source No. Trigger Source 1 Adverse reaction recorded de Wet7 2 Address of a residential facility Consensus 3 Home visit=de Wet de Wet7 4 >2 consults in a week Derived from de Wet (>3 consults)7 5 >12 consults per year Derived from de Wet (>10 consults)7 6 >3 consults with different GPs in a 3-month period Consensus 7 Predominant provider and nominated provider are different Consensus 8 No appointment & repeat Rx (repeat of previous medication) Consensus 9 No appointment & telephone Rx (medication not had previously) Consensus 10 Long-term medications and classifications are at variance Consensus 11 Diagnosis of cancer in the last 12 months Derived from de Wet (high priority READ code)7 12 Cessation of medications Singh6 13 >6 medications prescribed (at the same time) Consensus 14 Change of medications de Wet7 15 Reduction in medication dose de Wet7 16 Hospital discharge - including ED and day stay de Wet7 17 ED/A&M clinic after GP consult within 2 weeks derived from Singh6 and de Wet7 18 ED/A&M clinic after GP consult within 2 weeks prior to GP consult within 2 weeks de rived from Singh6 and de Wet7 19 ED/A&M clinic after nurse consult within 2 weeks derived from Singh6 and de Wet7 20 ED/A&M clinic prior to nurse consult within 2 weeks derived from Singh6 and de Wet7 21 Hospital admission with no GP consult within 6 months Singh and de Wet7 22 Attended outpatient clinic, including radiology, hospital clinics, physiotherapy & private specialists de Wet7 23 INR (5+) Singh6 24 Histology Consensus 25 Abnormal gynaecology cytology Consensus Lab results Source 26 eGFR <35 mL/min/1.73m2 derived from Singh6 27 TSH <0.03 on thyroxine) Singh6 28 Carbamazepine (Tegretol)>40 \u00b5mol/L Singh6 29 Digoxin (Lanoxin)>2 nmol/L Singh6 30 Phenytoin>80 \u00b5mol/L Singh6 31 Theophylline>110 \u00b5mol/L Singh6 32 Valproic acid>700 \u00b5mol/L Singh6 33 Lithium>1.5 mmol/L Consensus 34 Short-term admission to residential aged care facility Consensus 35 Death Singh6 36 Medication list not complete Consensus Rx=prescription. ED=Emergency department. A&M=Accident and medical. eGFR=Estimated glomerular filtration rate. INR=International normalised ratio. TSH=Thyroid stimulating hormone. Each record was then reviewed for the presence of any harm that was not related to the trigger. Harm was defined according to the Medication Error Index adopted by the National Coordinating Council for Medication Error Reporting and Prevention.8 Harm was classified according to the WHO National Coordinating Council for Medication Error Reporting.8 Following each session a reconciliation of findings between teams ensured consistency of interpretation of triggers and harm. If there was a difference between the two teams then a decision was made based on consensus. The analytic plan was first to measure the harm events associated with each trigger and calculate the sensitivity and specificity of each trigger. We then carried out logistic regression analyses, adjusting for sex, ethnicity and age to estimate the odds of harm associated with each trigger and with the 36 triggers combined. Using a consensus approach between members of the research team, triggers with the lowest specificity were then excluded and a refined trigger tool derived and tested for its ability to identify harm, using a further age-sex-ethnicity-adjusted logistic regression analysis. The study was reviewed and approved by the Northern X Ethics Committee (NTX/11/EXP/298). Results The records of 170 patients were analysed for both the presence of a defined trigger and the presence of harm - see Table 2 for demographics and Figure 1 for a flow chart of the analysis process and results. Thirteen patients had no trigger in their records. Table 2. Demography of patients whose records were reviewed Variables Male Female Total Age (years) <18 18-65 \u226565 24 37 17 17 55 20 41 92 37 Mori 44 41 85 Non-Mori 34 51 85 Total 78 92 170 A total of 1033 triggers were identified over a total of 40,030 days of follow-up in which 637 consultations were recorded. In these consultations, 44 harms were picked up by 62 triggers and 1 harm was not picked up by any triggers. All harms identified were medication related. Figure 1. Flowchart of analysis and results Table 3 lists triggers associated with harm. The rate of harm per consultation was 0.07 (95%CI 0.05-0.09) or 7 occurrences of harm per 100 consultations. The rate of harm per 100 patient years was 41 (95%CI 29-55). Of the 45 occurrences of harm: 34 (76%) were classified as Category E - temporary harm to the patient and required intervention; 8 (18%) were classified as Category F - temporary harm to the patient and required initial or prolonged hospitalisation; 2 (4%) were classified as Category G - permanent patient harm; and 1 (2%) were classified as Category I - patient death. The odds ratio of harm occurring using 36 triggers was 0.78 (95%CI 0.5-30) with a sensitivity of 0.98 and a specificity of 0.08. The refined primary care trigger tool included only 8 triggers: adverse drug reaction documented in the record, \u22652 consultations with a GP in the same practice in a week, cessation of medication, reduction in medication dose, \u22656 medications prescribed, attending the emergency department or an after hours provider within 2 weeks of having seen a GP, eGFR <35, and death. The odds ratio of harm occurring if one of the reduced set of triggers was present was 3.4 (95% confidence interval 1.7-7.1) when adjusted for age, sex and ethnicity. The sensitivity of this refined trigger tool was 0.81 and the specificity was 0.51. The odds ratio for harm occurring among male patients was 0.59 (0.32-1.10) and for Mori was 0.96 (0.48-1.93). The correlation coefficient for the refined primary care trigger tool, was 0.4 between the two groups of reviewers. Table 3. Number of consultations with a trigger and number (percentage) associated with harm) Trigger Number of consultatio