Unsafe patient care during health care delivery is one of the 10 leading causes of death and disability worldwide [1]. Worldwide, the frequency of adverse events in hospitalized patients has been estimated to range between 2.9% and 21.9%, with approximately half of these events considered to be preventable [2]. The most frequent patient safety events encountered in hospitalized patients include medication errors; health care–associated infections; unsafe surgical procedures, injections, and blood transfusions; venous thromboembolism; and diagnostic and radiation errors [3]. In the United States alone, medical errors result in health care costs of approximately US $20 billion and contribute to more than 250,000 deaths annually [4,5]. It is estimated that the global economic cost of unsafe care exceeds US $1 trillion annually [6]. However, the burden of unsafe care disproportionately affects low- and middle-income countries (LMICs), where it is responsible for 134 million adverse events and 2.6 million deaths annually [7]. In LMICs, poor-quality care now represents a greater barrier to reducing mortality than limited access to care, with approximately 60% of deaths associated with treatable conditions now resulting from deficiencies in the quality of care delivered [8].

Despite this long-standing disparity in burden, the vast majority of patient safety research has originated from high-income countries (HICs), underscoring a broader global trend in which scientific knowledge production is dominated by a small number of high-income nations [9,10]. However, in recent years, global awareness of the need to enhance patient safety standards has resulted in a modest increase in patient safety research in LMICs [11,12]. It is now widely recognized that local researchers are better able to establish health research priorities that are relevant to their populations; design contextually appropriate interventions; and effectively bridge the gap among evidence, policy, and implementation [13,14]. Therefore, a crucial first step in addressing unsafe care in LMICs is to generate locally derived evidence by identifying adverse events within health care settings, thereby determining the true burden and nature of the problem.

Worldwide, a range of methodologies and tools have been developed to detect health care–associated harm [15]. However, keeping resource constraints in LMICs in mind, it is important to use an adverse event detection methodology that is both accurate and feasible. Historically, most studies concerned with adverse events have used various medical record review methodologies that have greater sensitivity than other methods, including voluntary reporting [16,17]. The 2 most popular tools, derived from the Harvard Medical Practice Study and the Global Trigger Tool for Measuring Adverse Events, have been repeatedly validated and compared in studies conducted in HICs [2]. However, it is expected that, for a number of reasons, including rudimentary medical record documentation systems and a lack of available expertise, the results of those studies are not applicable to LMICs [7]. Furthermore, to the best of our knowledge, there are no real-world data comparing the 2 aforementioned tools exclusively from LMICs. Therefore, the primary aim of this external pilot study was to assess which of the 2 methodologies is most suitable for detecting adverse events in a tertiary health care setting in an LMIC. A secondary aim was to draw on the experience gained during the pilot to refine the selected methodology, including improving the data collection tools and workflow processes, in preparation for a larger study that will measure the incidence of adverse events in hospitalized patients.

The mean ages of patients in cohort A and cohort B were 39.2 (SD 27.7) and 39.8 (SD 28.4) years, respectively. Male individuals constituted 45% (45/100) of cohort A and 46% (46/100) of cohort B. The mean length of hospital stay for cohort A was 3.3 (SD 2.8) days compared to 3.5 (SD 3.4) days for cohort B. In cohort A, 36% (36/100) of the patients were admitted electively, whereas 39% (39/100) were admitted acutely. Conversely, in cohort B, 33% (33/100) of the patients were admitted electively, and 38% (38/100) were admitted acutely. Patient characteristics are summarized in Table 1.

A median time of 19 (IQR 13-27) minutes was required to review each file using tool 1A. A total of 114 positive triggers were found, resulting in the identification of 1 adverse event, which translates to an adverse event rate of 1 event per 100 hospital admissions. For tool 1B, a median time of 29 (IQR 19-57) minutes was required per file, and a total of 260 positive triggers were identified along with 30 adverse events, with 22% (22/100) of the patients experiencing at least one adverse event. Of these 30 events, 18 (60%) were considered preventable. The performance of both tools is compared in Table 1.

The most commonly identified trigger using tool 1A was >1 admission within 30 days, which was positive in a total of 27% (27/100) of the patients. Difference between diagnosis at admission and diagnosis at discharge was positive in 25% (25/100) of the patients, and National Early Warning Score 2 of ≥3 or Pediatric Early Warning Score of ≥2 on the day of discharge was positive in 21% (21/100) of the patients, representing the second and third most commonly identified triggers, respectively. The most frequently identified triggers in tool 1A are listed in Table 2.

Antiemetic administration, which occurred with 60% (60/100) of the patients, was the most frequently identified trigger using tool 1B, followed by transfusion of blood products, which was found in 19% (19/100) of the patients. The next most common trigger was time in the emergency department greater than 6 hours, which was found to be positive for a total of 18% (18/100) of the patients. The most commonly identified triggers using tool 1B are listed in Table 3.

Of a total of 30 events identified, surgical- or procedure-related complications were the most frequently identified type of adverse event, making up 40% (12/30) of all events identified. In this category, the most common event was surgery- or procedure-related hemorrhage, which was identified in 6% (6/100) of the patients and comprised 20% (6/30) of all events. Failure of diagnosis or treatment was the second most frequent category, making up 26.7% (8/30) of all events identified, followed by adverse medication events, which comprised 23.3% (7/30) of all events identified. The least common type of adverse event was health care–associated infections, only being identified in 1% (1/100) of the patients. Both tools 1A and 1B identified the health care–associated infection. The adverse events identified are summarized in Table 4.

This pilot study aimed to identify an appropriate methodology that could be used to accurately estimate the rate at which adverse events occur in hospitalized patients in LMICs. Using tool 1A, we obtained an adverse event rate of 1 event per 100 hospital admissions. However, using tool 1B, we obtained an adverse event rate of 30 events per 100 admissions, with a total of 22% (22/100) of the patients experiencing at least one adverse event. The most common type of adverse events identified were surgical- or procedure-related complications, with surgery- or procedure-related hemorrhage being the most frequently identified adverse event.

There is a lack of authoritative work regarding the incidence of adverse events in hospitalized patients in LMICs. However, the National Academies of Sciences, Engineering, and Medicine have estimated that 134 million adverse events occur across 531 million hospital admissions annually in LMICs, resulting in an adverse event rate of approximately 25 events per 100 hospital admissions [7]. However, the first methodology used, tool 1A, obtained an adverse event rate of 1 event per 100 hospital admissions. This is also in contrast to the work by Letaief et al [23], from which tool 1A was adapted. Their study found that, of 620 patient records reviewed, 62 (10%) contained evidence of at least one adverse event [23]. This suggests that tool 1A might be an inappropriate choice for estimating the rate of adverse events in our setting. We believe that this could be because the trigger list was originally developed and validated in HICs and, therefore, has limited applicability in our context. Differences in documentation practices, health care processes, and clinical workflows between HICs and LMICs may reduce the sensitivity and relevance of certain triggers. For example, variations in record-keeping systems, diagnostic resources available, and patterns of care delivery may lead to underdetection or misclassification of adverse events when using tools designed for high-resource settings [26].

The second methodology used, tool 1B, identified a total of 30 adverse events across 100 hospital admissions, which aligns more closely with regional estimates and, therefore, may be more suitable for use in our setting [7]. Moreover, Bates et al [27] conducted a similar study across 11 Boston hospitals using a trigger list that was also adapted from the Global Trigger Tool for Measuring Adverse Events. In their weighted sample of 2809 admissions, they identified at least one adverse event in 23.6% of the admissions, which is in line with our finding of 22% (22/100). This demonstrates that the tool may be applicable across diverse health care settings and is potentially suitable for future benchmarking efforts. It is worth noting that Bates et al [27] identified adverse medication events as the most common type of adverse event, accounting for 39% of all events detected. In contrast, surgical- or procedure-related complications were the most frequently identified type of adverse event in our study, comprising 40% (12/30) of all events recorded. We believe that this might reflect the difference between the quality of surgical care provided in LMICs compared with HICs [28]. Lack of access to timely surgical intervention, low surgical case volumes, and a lack of specialists are often cited as the main contributors to unsafe surgical care in LMICs [28].

An important consideration with tool 1B is that it generates a substantial number of positive triggers, many of which do not ultimately correspond to true adverse events yet still necessitate a detailed review of medical records. However, this broad approach may be preferable within nascent patient safety systems, especially in LMICs. By casting a wider net, tool 1B reduces the risk of missing relevant events and allows organizations to assess and prioritize identified events based on relevance and impact. Additionally, analyzing and addressing these triggers through a system-based approach may promote wider organizational improvements, yielding safety benefits and fostering resilience that may not be possible by relying solely on reactive approaches to safety issues [29].

We would like to acknowledge the salient limitations of this methodology. Retrospective chart reviews are limited by the fact that they only use data that were previously collected during the delivery of routine care and, therefore, cannot capture the full range of harm experienced by patients. For example, any adverse events that manifest after discharge but do not result in further patient encounters will be missed. Moreover, this issue is further exacerbated by poor documentation culture and practices in LMICs, which result in missing information, difficulty in interpreting information (such as due to the use of nonapproved abbreviations), and questionable authenticity of the information recorded. However, despite these limitations, retrospective chart reviews remain the most appropriate choice for our setting as they are relatively inexpensive; use existing data; and allow us to capture events that manifest after short latency periods, such as postoperative complications. Finally, we would like to acknowledge that tools 1A and 1B were not applied to the same set of medical records. However, as no adverse events were detected in the records involved, any potential impact on the findings is expected to be minimal.

In conclusion, this pilot study enabled us to identify and select well-defined, validated tools for use in our larger study, which will allow for reliable measurement of the incidence of adverse events among hospitalized patients. The methodological and operational lessons learned will directly inform our larger study and future capacity-building efforts in data quality, training, and surveillance systems [30]. Furthermore, by building local capacity for adverse event detection, we contribute to broader efforts to strengthen health systems by identifying priority areas for quality improvement, guiding resource allocation, and informing the design of context-appropriate interventions to reduce preventable harm. At the policy level, evidence generated through such locally led initiatives can support the development of national patient safety strategies, establish frameworks for data-driven decision-making, and integrate patient safety indicators into routine health system performance monitoring.