Trigger-based Patient Safety Method Finds High Rate of Preventable Harm in Hospitals – New Harvard Study Reveals

Dr. David Bates & Colleagues Publish Harvard Hospitals’ Safety Study in The New England Journal of Medicine

While the 1991 Harvard Medical Practice Study (HMPS) has served as an oft cited milestone for measuring patient harm for decades, Dr. David Bates, Dr. Elizabeth Mort, and co-authors from Mass Gen Brigham and affiliates sought to evaluate the current state of patient safety in light of many improvements in the field since the publication of the HMPS in 1991, and studies released since then.  In the new study titled, “The Safety of Inpatient Health Care”, The authors pointed out that since that HMPS study, “more efficient and reliable approaches have been established for the identification of adverse events, including the use of ‘triggers’”^[1]Bates, David W et al. “The Safety of Inpatient Health Care.” The New England Journal of Medicine vol. 388,2 (2023): 143.– the methodology that lies at the core of Pascal’s comprehensive Virtual Patient Safety (VPS) solution.  The authors used Pascal’s trigger-based method to identify adverse events across 11 Massachusetts hospitals using 2018 data.

To summarize the remarkable findings, the CRICO-funded^[2]CRICO is the acronym for the Controlled Risk Insurance Company and the Risk Management Foundation of the Harvard Medical Institutions, NEJM-published research identified one adverse event in every four admissions, and approximately a quarter of those were preventable.

Dr. Don Berwick weighed in on the findings in an accompanying editorial.

As quoted further below, the co-authors suggest that the future of patient safety is identifying adverse events with triggers and advanced analytics, standardizing the collection and use of such data, and sharing successful improvement strategies derived from that data. 

Results Summary

• Across 2809 admissions, the trigger-based adverse event detection method identified at least one adverse event in 23.6% of those patients;
• Of 23.6% of adverse events, 22.7% were judged to be preventable;
• Of 23.6% of adverse events, 32.3% were judged to be serious harm (requiring substantial intervention or prolonged recovery);
• A preventable adverse event occurred in 6.8% of all admissions;
• Serious harm occurred in 1% of all admissions;
• Of the seven deaths, one was deemed preventable; 
• The categories of adverse events and corresponding percentages of all harm were:  adverse drug events (39%); surgical or procedural events (30.4%); patient-care events (i.e. associated with nursing care) (15%); and health care-associated infections (11.9%); and 
• The rates of preventable adverse events ranged across the 11 hospitals from 0.9 to 10.9 events per 100 admissions.
• The mean length of stay for admissions during which an adverse event occurred as more than 2x as long as for those admissions without an adverse event:  9.3 days versus 4.2 days

Research Rationale

The co-authors point out additional facts providing rationale for this study:

• Major efforts to measure such as by the HHS Office of Inspector General, while showing a “moderate serial decrease in the incidence of health care-associated infections, many key safety domains lack metrics that can be easily measured by organizations to routinely track adverse events and assess progress in improving safety.” ^[3]Bates, David W et al. “The Safety of Inpatient Health Care.” The New England journal of medicine vol. 388,2 (2023): 143. One example of this is in the area of adverse drug events, where one study showed measured incidence was 20 times higher than identified through voluntary reporting.
• The advent of data liquidity through the implementation of EHRs is also a development since HPMS, having moved from being rare in 1991 to being in routine use now.
• While this study only covers inpatient, a substantial portion of care today has moved from the inpatient setting to outpatient (perhaps a follow-on.)

Health systems do not measure adverse event outcomes using EHR data with clinical validation (AE Outcomes).  Consequently, the data do not exist at most health systems for the questions – such as the first posed above – with evidence.  Indeed, the field historically has relied on “proxy” adverse event outcomes data, i.e. those which are not sufficiently fine-grained, patient-specific, and validated using clinical data.

Practical Implications

The co-authors make a number of key points in their discussion:

1. Harm remains a problem but is preventable.  Adverse events remain common and are preventable nearly one fourth of the time.
2. Estimated rates are likely conservative.  Although this study tracked adverse events that were present on admission (POA), the methodology excluded these adverse events from rate calculations, making them more conservative. 
3. Included temporary harm as an adverse event.  Unlike in the HMPS but like the Landrigan et al study, temporary harm is considered an adverse event. 
4. Applying machine learning to diagnostic events.  Co-authors see machine learning as being able to assist particularly in the identification of diagnostic errors (because the study’s trigger-based approach was limited in being efficient, reliable, and continuous – unlike Pascal’s trigger-based VPS solution).
5. Voluntary event reporting undercounts harm and is inadequate.  Many U.S. hospitals rely solely on voluntary event reporting, which “results in substantial undercounting and, in some cases, misleading reports of zero harm.”^[4]Ibid: 152

Key Takeaways

The co-authors suggest a number of takeaways:

1. Technology.  Citing evidence generated by a Pascal Risk Trigger Monitor (RTM) implementation with a large health system): “Identification of adverse events in EHRs in the future will probably be performed by means of computerization of triggers and also through leveraging of artificial intelligence.” ^[5]Ibid, p 152
   Pascal VPS supports this path for health systems.
2.  Standardization.  This work needs to be operationalized with the ability to scale: “Measuring adverse events in a reliable and efficient way and developing standard approaches to the identification of and focus on preventable adverse events are critical to supporting persons charged with improving safety.” ^[6]Ibid, p. 152
   Pascal VPS enables health systems to monitor, measure, and manage with standardized data, process, and technology.
3. Improvement.  Supported by safety culture and strong leadership, it is possible to improve (i.e. reduce both the frequency and severity of preventable harm) with advanced monitoring, reporting, data analytics:  “This finding [variability in adverse event rates] suggests that if hospitals had data that were more reliable and more routinely collected, it is possible that monitoring could be improved, adverse events rates could be reduced, and improvement strategies could be shared through careful study of interventions.” ^[7]Ibid, p. 152
   Pascal’s VPS clients have seen these improvements first-hand in both all-cause and specific harm reduction.