Accuracy of triggers in the identification of adverse drug events in hospitalized elderly

Objective: Evaluate trigger accuracy for identifying adverse drug events (ADEs) in hospitalised elderly. Methods: Two hundred patients ≥60 years old from a medical clinic within a private hospital were followed-up. For ADE identification the adapted Global Trigger Tool tracker methodology was used. Causality was determined using the Naranjo Algorithm. Results: Of the 200 elderly patients included in the study, 106 were females (53%), the average age was 79 years, and the average length of hospital stay was approximately ± 10 days. Selected triggers were identified 1,457 times. The group of triggers with the best performance regarding its analysis accuracy was evolution triggers, with sensitivity of 69% and positive predictive value of 68%. In the individual performance analysis, the evolution tracker allergy, allergic reaction, pruritus achieved 100% performance for both sensitivity and positive predictive value. A total of 165 ADs were identified. Of these, 18% were phlebitis and 16% were hypoglycaemia. Drugs associated with ADE Research, Society and Development, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 2 included insulin (15%) and Clarithromycin (9%). Conclusion: The triggering methodology has been effective for identifying ADEs. In addition, determination best trigger for constructing an ADE identification tool for hospitalised elderly was performed.


Introduction
The World Health Organisation (WHO) (2002) has defined an adverse event (AE) as any incident that results in unintended damage arising from care and unrelated to the natural evolution of a patient's disease. The occurrence of an adverse drug event (ADE) is one of three main explanations for AE (Mendes, Martins, Rozenfeld, & Travassos, 2009). For this reason, the WHO (2017) launched the third global challenge for patient safety: harmless medication to reduce the prevalence of ADE in the world.
The use of medicines is essential in health care. Medicines are used to prevent and treat several pathologies, and are responsible for enhancing quality of life (Gomes, Silva, & Galvão, 2017). Despite this, there are risks related to this technology that must be considered by the health team, institutions and health authorities in order to create barriers that prevent adverse events. ADE are associated with damage to patient health, increased length of hospital stay and increased health service costs (Almeida, Castro, & Caldas, 2011).
In the United States of America (USA), adverse events are the third leading cause of death, and at less one medication error occurs per day in each hospitalised patient (Makary & Daniel, 2016). These errors, despite being preventable, can cause AEs. Thus, hospitalised patients are particularly susceptible to the occurrence of ADE (Charles, 2010).
Some factors predispose patients to ADE including extremes of age, clinical severity, comorbidities and the use of polytherapy. These factors characterise geriatric patients as an at risk group for the occurrence of ADEs (Silva, Ribeiro, Klein, & Acurcio, 2012). Physiological Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 4 changes that occur throughout the aging process also interfere in pharmacokinetic and pharmacodynamic processes, and predispose this age group to toxicity, drug interactions and adverse events. In addition, the geriatric population is one of the most medicalised group in society due to self-medication, a high incidence of chronic degenerative diseases and progressively decreasing functional capacity (Carvalho et al., 2012;Pizzol et al, 2012). Some strategies of active searching have been used to identify and prevent ADE.
Among these, we highlight triggers, which are features of the medical records of patients that may indicate the presence of ADEs. Various triggers have been described in the literature.
The Institute developed the "Global Trigger Tool" (GTT) for Healthcare Improvement (IHI) and uses retrospective analysis of medical records to monitor AEs. Health institutions for implementing safer practices and improving procedures use this data. The prospective use of this tool produced variable results for AE identification (Griffin & Resar, 2009). This study aims to evaluate the accuracy of use of triggers in the identification of adverse drug events (ADEs) in hospitalised elderly patients.

Study setting and subjects
This observational and prospective study was conducted from August to December of 2018, in a 150-bed private hospital in Belém-Pa, Brazil. The Ethics and Research Committee of the Federal University of Pará (code number 97957518.4.0000.00186) approved this research.
Inclusion criteria were as follows: patients aged ≥60 years old, admitted to the hospital's medical clinic during the mentioned period, with a hospital stay longer than 48 hours. Where the patient and/or legal guardian agreed to participate in the study by signing the informed consent form. Exclusion criteria included all patients with communication difficulties, patients in contact isolation, or patients with no clinical outcome. Throughout the study period, there were 461 hospitalisations. Of these, 90 were not elderly and 39 met exclusion criteria.
A sample calculation determined that 179 of the total 332 eligible elderly patients would be required to produce 95% confidence. Data presented in this work correspond to 200 elderly people monitored and randomly selected using the program BioStat®. In this work, the WHO (2009) definition of ADE was used, which states that ADE is defined as "any Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 5 incident resulting from the process of using drugs that result in damage or injury to the patient, including adverse drug reactions and medication errors".

Triggers
Was used Triggers from the medication module of the Global Trigger Tool methodology, as shown in Chart 1, proposed by Griffin & Resar (2009). In addition, Chart 2 describes clinical changes identified in a study by Silva (2017), which were adapted as triggers in the present study. The triggers that correspond to medicines were considered present when they were both prescribed and administered (as noted by the nursing team). Chart 1. IHI trigger for the "Medicines" module. Chart 1. IHI triggers from the medication module of Global Trigger Tool methodology.

Multiprofessional Evolution
• Rash and/or itching and/or allergy and/or allergic reaction.
Registration of one of the triggers in the multiprofessional evolution.
Heart rate recording ≥100bpm associated with drugs that cause tachycardia.

Multiprofessional Evolution
• Phlebitis Registration of phlebitis associated with infusion of irritating and/or vesicant medication.

Data Collection
All patient data were assessed by two independent (pharmacists) reviewers. If there was doubt and/or disagreement regarding the occurrence of an ADE, a third reviewer (physician) was consulted. Reviewer training was carried out using a training programme provided by IHI, which taught reviewers to use the Global Trigger Tool. The frequently asked questions section also was used to answer questions that arose during training. A pilot study was carried out with 50 patients to make necessary adjustments to the data collection tool.
Results obtained during the pilot study were not considered in this study.
Data collection involved the following four stages. Stage 1 involved the identification of hospitalised elderly patients. The Hospital Census Report of the MV SOUL® Management System was used to identify patients. This report allowed researchers to access a list hospitalised patients that included the following patient data: date of birth, name, attending physician, medical care number, and date of hospitalisation. Subsequently, the inclusion and Research, Society and Development, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 9 exclusion criteria were applied to define which elderly patients could be considered for participation in the research.
Stage 2 involved the acquisition of patient consent. After 48 hours of hospitalisation, a bed visit to each patient was made, and research objectives were explained. At this time, patients were asked to provide informed consent.
In Stage 3, patients were monitored. At the same time, patients answered a questionnaire during an interview. Information collected included life and physiological habits and continuous medication use. The patients were monitored daily during their entire period of hospitalisation using an MV PEP® system, which allowed researchers to view prescriptions, laboratory tests and access team evolutions to search for triggers. Data was recorded on a Google form.
Stage 4 involved determining causality of trigger. After recording information determined during Stage 3, strength of the cause of each ADE was determined using clinical evidence provided in the event causality chain algorithm, which was previously described in Naranjo et al. (1981). A duration not exceeding 20 minutes was used, as recommended by the IHI. All the technical information regarding medicine was determined using the Micromedex Solution® System.

Statistical Analysis
The determination of the accuracy of triggers in the identification of ADE was performed using measurements of sensitivity, positive predictive value (PPV), specificity and negative predictive value (NPV). These values were set individually for each tracker and for trigger groups. Sensitivity was defined as the proportion of patients with an ADE that were identified by the tracker and patients with an ADE that was not identified by the tracker. PPV was defined as the proportion of ADEs identified by the tracker and the number of unidentified patients who experienced ADEs. Specificity was defined as the proportion between patients who did not have ADE and in the presence of any tracker compared to patients who had the tracker, but did not have ADE. While NPV was the proportion among patients who had neither ADE nor the presence of a tracker for patients who presented ADE, but was not identified by a tracker. Statistical analyses were performed using BioStat® Software and all analyses, parameters were considered significantly different when p < 0.05.

Results
The average age of the 200 patients selected was 79 years old, and 106 (53%) were female. The average length of each patient's hospital stay was 10 days, and patients who had ADEs had an average stay of 12 days, compared the 6-day average that was determined for patients who did not experience an ADE. Eighty-two patients (41%) were hospitalised due to an infectious condition. The non-specific urinary tract infection (UTI) was the most prevalent International Classification of Disease category in patients assessed, and occurred in 15% Regarding comorbidities, 98% (196) of the elderly patients considered had at less one comorbidity. Among the most prevalent diseases observed, we highlight: systemic arterial hypertension (SAH), diabetes mellitus, and malignant neoplasia. Due to the high incidence of comorbidities, 91% (182) of patients reported using at least one medication continuously.
During hospitalisation, an average of 14 prescriptions, 162 doses of medication, and 14 different active ingredients were provided per patient. The distribution of pharmacotherapeutic risk was as follows: 65%, 29%, and 6% of elderly patients were classified to be at moderate, high, or low risk, respectively.

Trigger Identification Accuracy
In this study, the triggers occurred 1.457 times ( 7.2) in 99.5% (199) of the elderly hospitalised patients considered. Of these, 13.7% (200) of related triggers corresponded to 76% (125/165) of ADEs identified the survey, and other ADEs were identified without a trigger.
The absence of triggers was observed in only one patient throughout the duration of hospitalisation. Selected triggers occurred 1,457 (one thousand four hundred and fifty-seven) times, and an average of 7.2 triggers were observed per patient. Of these, 200 (13.72%) were related to the occurrence of an ADE. Sensitivity and PPV were established both for groups of triggers and for each trigger individually. In individual analyses of triggers, sensitivity varied between 0.92 and 1.0 and NPV varied between 0.32 and 1.0. The lowest NPVs were associated with the following triggers: blood glucose > 140mg/dL (capillary or serum, 0.32); Blood pressure < 140/90mmHg (0.46), Heart rate > than 100 bpm (0.72). Other triggers had NPVs over 0.92.
Triggers were placed in the following categories: medication, evolution, laboratory tests and vital signs. The following triggers did not occur in patients during the study period: medicines (cetirizine, dexchlorpheniramine, flumazenil, naloxone and prednisolone); evolution (arrhythmia and tremor); laboratory tests (international normalised ratio (INR) > 6, positive test for Clostridium difficile; and haematocrit > 30% or absolute reduction in haematocrit ≤ 10%. Meanwhile, the all vital signs trigger was identified at least once.
In addition, the following identified triggers were not associated with ADEs, which made it impossible to determine Sensitivity and PPV: medicines (bromopride, dexamethasone, dimenhydrate, methylprednisolone, metoclopramide, and Vitamin K); evolution (insomnia, hypertension, and dizziness); laboratory tests (creatine increase greater than 50% of the baseline for more than 7 days; time of activated partial thromboplastin ( (aTTP) > 100 s; increase of bilirubin 2x the elevation of transaminases, without elevation of alkaline phosphatase and INR ≥ 1.5); vital signs (blood pressure < 140/90 mmHg).

Medication module trigger prediction accuracy
An individual analysis of the medication group revealed sensitivity values between 0.17 and 0.67 and PPVs between 0.02 and 1.0. The best parameter for identifying ADE was the diphenhydramine trigger (sensitivity, 50%; PPV, 100%). Despite having a sensitivity value of 67%, hydrocortisone had a low PPV (10%). Table 1 includes values other sensitivity and PPV measurements. Subtitle: S (sensitivity), E (specificity), PPV (positive predictive value) and NPV (negative predictive value) (-) It was not possible to determine sensitivity and PPV because, according to the researchers' assessment, although the triggers were identified during hospitalisation, they were not related to an ADE. Source: Authors.

Evolution module trigger prediction accuracy
Trigger sensitivity values and PPVs, which were 0.5-1.0 and 0.05-1.0, respectively, are described in Table 2. The evolution trigger with the best performance was the allergy/allergic reaction/skin rash, pruritus trigger, which had a sensitivity value of 100% and a PPV of 100%. The abrupt interruption of medication trigger followed with a sensitivity value of 100% and a PPV 61%.
Research, Society and Development, v. 9, n. 12, e33491211135, 2020 (CC BY 4. Subtitle: S (sensitivity), E (specificity), PPV (positive predictive value) and NPV (negative predictive value) (-) It was not possible to determine sensitivity and PPV because, according to the researchers' assessment, although the triggers were identified during hospitalisation, they were not related to an ADE. Source: Authors.

Laboratory Test module trigger prediction accuracy
The laboratory tests trigger (Table 3) predicted with the greatest degree of accuracy was the increase in creatinine by 0.3 mg/dL (at least two days) trigger, which had a sensitivity value of 67% and a PPV of 59%, while the creatinine 2x above basal had a sensitivity value of 13% and a PPV of 40%. Subtitle: S (sensitivity), E (specificity), PPV (positive predictive value) and NPV (negative predictive value) (-) It was not possible to determine sensitivity and PPV because, according to the researchers' assessment, although the triggers were identified during hospitalisation, they were not related to an ADE. Source: Authors.

Vital Sign module trigger prediction accuracy
The accuracy analysis of the vital sign trigger is shown in Table 4. Sensitivity varied from 0.24 to 1.0 and the PPV varied from 0.02 to 0.94. The trigger with the best performance was glycaemia > 50 mg/dL (capillary or serum), which had a sensitivity of 68% and a PPV of 94%. And although the trigger glycaemia > 140mg/dL (capillary or serum) had the best sensitivity (93%) among the evolution triggers, its PPV was only 12%. Subtitle: S (sensitivity), E (specificity), PPV (positive predictive value) and NPV (negative predictive value) (-) It was not possible to determine sensitivity and PPV because, according to the researchers' assessment, although the triggers were identified during hospitalisation, they were not related to an ADE. Source: Authors.

Analysis of Trigger Groups
A predictive accuracy analysis of trigger groups is described in Table 5. In this study, an accuracy analysis of triggers revealed that the group of drugs trigger had the highest sensitivity (72%), but lowest PPV (9%). Therefore, the probability of the drug tracker being related to an ADE was determined to be low. The laboratory tests trigger had moderate sensitivity and PPV (62% and 31%, respectively), but had the best specificity (E: 89%) to negative negative predictive value NPV (97%) ratio, meaning that the tracker absence correlated with event absence. However, the evolution tracker had the best ratio between sensitivity (69%) and PPV (68%). Therefore, according to the multidisciplinary team, evolution data were optimal for identifying triggers of ADE.  Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 16 this category increased the length of the patient's hospital stay. Finally, 3 (2%) events were associated with patient death. One was likely due to a hypersensitivity reaction to iodinated contrast, and the other two were associated with antihypertensive drugs that, due to the severity of the patients' conditions, contributed indirectly to their deaths.
There was no correlation observed between the occurrence of ADE and gender.
Variables that best correlated with the occurrence of ADE (> 95% CI), according to Pearson's correlation, were: length of hospitalisation (r² = 0.41), total drug doses used (r² = 0.33) and number of active ingredients (r² = 0.30). A selected linear regression was applied to define factors associated with length of stay. The number of ADEs and the number of active ingredients prescribed were identified, which had r² values of 2.78 and 0.55, respectively.
In this study, the difference between the hospital stay length in patients with ADE(s) and those who did not experience an ADE was about 6 days.

Discussion
The geriatric patient experiences physiological changes that result from the aging process, and tends to have a greater number of comorbidities associated with acute or chronic health conditions. These characteristics of geriatric patients contribute to their increased risk of hospitalisation (Gon & Kending, 2016. Nunes et al., 2017. Queiroz, Oliveira, Araújo, & Reis, 2016. The national health research in Brazil, states that hospitalisation event is more frequent, with a longer hospital stay and consequently greater cost, in elderly patients (Malta et al., 2017).
The elevated prevalence of elderly men may have been a result of the gender's increased propensity to neglect their health, their relatively reduced use of child and secondary healthcare services, their elevated risk of clinical decompensation and, consequently, elevated risk of death versus females (Almeida, Mafra, Silva, & Kanso, 2015;Chaimowicz, 2009). In addition, in the present study, 50.6% of the studied hospitalisations involved elderly individuals older than 80 years old, of which 65.9% (66) were females. According to some authors, elderly women have a life expectancy that is 10 years longer than men. This justifies the female predominance regarding individuals hospitalised ≥ 80 years old (Anderson iet al., 1998;Robb, Elizabeth, Ashika, & Richard, 2017). Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 Some authors have correlated ADEs with age, length of stay, female gender and presence of multiple diseases (Rahn, 2008;Robb et al., 2017;Seddon, Cameron, Young, Maharaj, & Miller, 2013). In this study, 60% of patients who had ADE (identified by trigger or not) were female. However, similar to another study developed in Brazil (Almeida et al., 2011), there was no statistical association between ADE and gender.
The average length of hospital stay in this study was determined to be 10 days, which differed from the average duration of hospitalisation of 6 and 16 days, which were reported by Rodrigues et al (2017) and Santos (2007) respectively. This difference may be due to an early hospital discharge protocol initiated at the institution where this study was conducted. Since 86% of patients were discharged as a result of clinical improvement, and only 5% were discharged as a result of being completely cured. The early discharge planning scheme implemented by the hospital multidisciplinary team reduced length of hospital stay, iatrogenic occurrence, functional loss, poor outcomes after discharge, hospital readmissions and mortality (Varallo et al., (2017).
Infections were responsible for 41% of hospitalizations, however, in a study carried out by Queiroz et al., (2016) in the ICU of a hospital in Salvador, only 15.5% of hospitalizations were related to infection. The high rate of hospitalisation due to infections found in this study may be a result of the institutionalised sepsis protocol, which facilitates the identification and management of infectious diseases.
Of total reported infections, UTI accounted for 15% of hospitalisations. In elderly patients, UTI are common and typically affect 20% of elderly women and 10% of men. UTI is a major cause of sepsis in hospitalised patients. The high prevalence of UTI in elderly is likely a result of physiological changes, functional and self-care capacity loss, low tolerance to therapeutic procedures, difficulty walking and challenges with use of urinary devices (Silva, Pedreira, Santos, Barros, & David, 2018;Thiago, 2010;Veronese, 2005).
The high occurrence of comorbidities among the elderly (98%), especially systemic arterial hypertension (SAH), Diabetes Mellitus (DM) and Malignant Neoplasm (NM), was also noted by other authors, who both reported a rate of comorbidities in the elderly of 91% (Caldas et al., 2015;Motta, 2002). The high rate of comorbidities observed is also associated with the increased number of drugs used continuously, prescriptions, active ingredients prescribed, and dispensed drugs in elderly individuals, which was also observed in this study. This is reflected in the pharmacotherapeutic risk classification in patients at moderate and high risk for developing drug-related problems, and the almost nonexistent decrease in patient Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 18 risk during hospitalisation. This was observed since the number of medications prescribed and age itself were determining factors for scores adopted in the studies.
Regarding triggers, after searching literature databases, it became apparent that most articles did not use the GTT methodology exclusively. Instead, researchers combined the GTT methodology with others and/or triggers to identify triggers with greater specificity. In the present study, triggers occurred in 99.5% of hospitalised elderly patients. Studies in a hospital in the Federal District (Almeida et al, 2001) and another in a tertiary hospital in Rio de Janeiro (Rozenfeld, Giordani, & Coelho, 2013) determined that 9.7% and 70% of the assessed population, respectively, presented a trigger. A study that applied a variation of the GTT methodology and assessed eight US hospitals, determined 87.7% of patients considered had a trigger (Kennerly et al., 2013). In a study carried out in Brazil with elderly admitted to an emergency room, 38 triggers were found in 278 records analysed, eight ADEs were identified.
The high number of patients who had trigger in this study, and the number of triggers related to ADE, are probably a result of the modification rate of the GTT method proposed by the IHI, which included additional triggers since a study from New Zealand concluded that the rate of GTT tool modification is a determining factor in the ability of the methodology to identify ADE (Seddon et al., 2013). In addition, in retrospective studies, the absence of complete data in medical records has been shown to be a determining factor, as was observed in a study developed in an surgery center, which considered 207 patients. Of the total number of patients considered, 25% did not have complete data in their medical records and 17% did not have available laboratory data. As a result, triggers were associated with only seven ADEs. The authors suggested the need to propose new triggers (Franklin, Birch, Sachter, & Barber, 2010).
Most studies analysed trigger accuracy via exclusively PPV to determine trigger performance. The formula used to evaluate accuracy remains controversial. Some authors consider the occurrence of a tracker per patient, but others consider trigger presence per day.
In this study, a low predictive value was associated with the high prevalence of comorbidities Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 such as SAH and DM, and difficulty establishing a causal relationship between triggers and events occurred due to the clinical complexity of elderly patients.
In this study, the trigger sensitivity varied between 17 and 100% and the PPV varied between 2 and 100%. The PPV was similar to the results of the study by Rozenfeld et al., (2013) and Matlow et al., (2011), who reported values of 12-100% and 0-83.3%, respectively. In the last article trigger analysis was probably not excluded, but no ADE were associated with triggers. In a prospective study conducted in a hospital medical and surgical clinic in Brazil, PPV and NPV were calculated according to days on which the triggers occurred. In the study, trigger sensitivity ranged from 0.3 to 11.8% and PPV ranged from 1.2 to 27.3%, while specificity and NPV were greater than 86% (Silva, 2017). Values below those found in this study were also observed in a different study in which PPV varied between 0-0,67% (Carnevalli, et al., 2013).
Regarding drug triggers, hydrocortisone, diphenhydramine and fexofenadine have been reported to be the best triggers, with sensitivity values of 57%, 14% and 14%, respectively (Almeida et al., 2011). Variation between studies can occur, due to service profile itself, and protocols for medicine use established throughout study sites.
The trigger creatinine increase by 50% from baseline (for more than 7 days) was associated with kidney damage, there was only one occurrence of the tracker, which was not related to an ADE but instead to pre-existing kidney damage. The low occurrence of the tracker may be associated with the 7-day period needed to define the trigger. In the 44% (88) of patients who had a hospital stay of less than 7 days, it was not possible to identify the trigger. Further, it must be considered that the creatinine baseline does not always occur on the first day of hospitalisation.
In this study, although glycaemia > 50mg/dL (capillary or serum) and the progress tracker hypoglycaemia (sensitivity, 50%; PPV, 92%) performed well, mistakes related to filling out vital sign data in the patient's electronic medical record were observed, and despite blood glucose records appearing to be greater than 70mg/dL in the system, when investigated, 50% glucose was dispensed and administered to 100% of the patients who presented with hypoglycaemia. Thus, when we analysed factors that contributed to the events, all were preventable by performing simple measures such as insulin administration, training, and maintaining compliance with protocols for safe use of medicines that were already instituted within hospitals. These data are in accordance with findings reported by another study that evaluated the occurrence of ADE in which 19% (8) of the events hypoglycaemic events were associated with incorrect infusion practices (administration of an incorrect volume of insulin), Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 20 and two cases of abuse by hypoglycaemic events were classified as preventable (Nilsson, Pih, Tagsjo, & Ericsson, 2012).
The performance of each tracker should be analysed in order to identify the best trigger for ADEs in a particular context and, possibly, subsidise the creation of a tool adapted to carry out pharmacovigilance. The number of ADEs found in this study (165) was similar to the number Passarelli (2005) identified in a study that involved 186 hospitalised elderly patients and identified 199 ADE. Another study that also evaluated the use of GTT in a hospital in Brazil identified hypotension and constipation as principal adverse events, and the ADEs were responsible for 21.2 and 18.3% of all identified ADEs, respectively (Silva, 2017).
The recommendation of the Infusion Nurses Society is that phlebitis occurs at a rate >5%. The incidence of this ADE may vary according to its means of identification (medical records, active search, spontaneous notification), or protocols for the infusion of drugs established at relevant institutions (Infusion Nurses Society, 2006). In this study, of the 30 total phlebitis cases observed, 60% were not identified by the phlebitis evolution tracker, since the event was not described in the medical records and was identified via a bedside pharmaceutical visit and/or via the questioning of health professionals involved in patient care. Of the total phlebitis events, 11 were related to medication errors such as incorrect dilution or incorrect infusion time, even though pharmacists evaluated all prescriptions prior to their use, and the system used at the hospital already included information regarding suitable dilutions and infusion times to guide the prescriber. However, some prescribers changed information manually and, in other cases, medications was prescribed correctly, but their dilution, reconstitution and infusion time were incorrectly carried out by the nursing staff.
Further, as the study progressed, an association between contusion gel dispensing and the occurrence of ADE was observed. When we analysed medication dispensed, 29 patients (96.2%) used the gel, which was a possible tracker of phlebitis within the institution, and had an advantage of dispensal data being available in the pharmacy sector. The ADEs found, including phlebitis, were mostly associated with the use of antimicrobials. These results are in accordance with other studies, which also reported anti-infectives (Almeida et al., 2011), antidiabetics (Shehab et al., 2016) and cardiovascular agents (Rozenfeld et al., 2013) as main causes of ADE. Regarding causality, in accordance with this work, a prior study revealed that 71% of ADE were classified as possible and 29% were classified as probable (Almeida et al, 2011). A systematic review with meta-analysis by Hakkarainen (2012), which evaluated Research, Society andDevelopment, v. 9, n. 12, e33491211135, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i12.11135 21 preventability, revealed that at least half of all ADEs were preventable. This was in accordance with findings reported here, which showed that 62% of events were preventable.
In the Rozenfeld et al., (2013) study, of the total of adverse events identified, 82% contributed to or caused temporary damage requiring intervention, and 6.0% may have contributed to the death of patients. Rozich, Haraden & Resar (2003) reported that 79.9% (219) of events resulted in temporary damage to patients, value close to that which was in this work, and 2% of the events were associated with patient death.
The presence of a comorbidity (13%), number of active ingredients used (30%) and total number of dispensed drugs (33%) were associated with ADE occurrence. This is in accordance with published findings that reported that polytherapy, and comorbidities were factors related to ADEs ] . Multiple linear regression identified comorbidities, number of active ingredients and number of ADEs as factors responsible for 56% of hospital stay duration variation. In isolation, the occurrence of an ADE was responsible for 41% of hospital stay variation. Rozenfeld et al., (2013) established that the length of hospital stays for patients with ADE was, on average, 35 days, and 11 days for patients without ADEs. The patient's length of hospital stay is associated with an increased risk of care events and an elevated cost of health services.
Limitations of this study include the authors' interpretation of the possibility of the prevention of an event, since the line between what is avoidable and unavoidable is subtle and sometimes subjective (Nilsson et al.,2012). We also highlight the quality of the electronic record used. Even as a prospective study, it was possible to identify flaws regarding filling out patient data. In addition, patients are continually monitored by clinical pharmacists at the hospital, and this study did not estimate the number of ADEs that were prevented by these professionals.

Conclusions
The trigger tool proved to be effective for identifying ADEs. The performance was similar to that which was reported by other authors who estimated that triggers can be used to identify up to 20 times more events. In fact, only seven events were noted by the multiprofessional team during the study period. The best performing triggers were medication (diphenhydramine and promethazine); evolution (allergy/allergic reaction/skin rash, pruritus; hypoglycaemia; abrupt stop of medication); laboratory tests (increase of creatinine by 0.3 mg/dL); vital signs (blood glucose < 50mg/dL (capillary or serum), blood pressure < 90/60mmHg), dispensal of medication for contusion, and 50% glucose (if necessary). Event identification and the analysis of its cause is necessary to characterize health programmes and to identify failures in health care and follow through indicators, and improve protocols that are implemented to promote the safety of care, and the rational and safe use of medicines. All events identified in this survey were reported to the hospital's risk management department so that individual analyses of each event using the London Protocol could be performed.

Data availability statement
Data that support the findings of this study are available from the corresponding author upon request. The data are not publicly available due to privacy and/or ethical restrictions.