Sentinel Event Rate

Sentinel Event Rate in Patient Safety Measurement

A sentinel event rate tells a hospital or health system how often its gravest patient-safety failures occur relative to the care activity that could have produced them, such as surgical procedures. Such events are rare, so the rate rests on a handful of cases and its uncertainty is wide. Most events carry high costs and large health losses, so a credible rate is the starting point for valuing prevention. The reported count also depends on whether staff detect and report events, so a carelessly built rate can misdirect safety spending.

What counts as a sentinel event

The rate inherits its meaning from the event definition, set by the body whose scheme is used. The Joint Commission, which accredits healthcare organisations in the United States, defines a sentinel event as a patient safety event, not primarily related to the natural course of the patient's illness or underlying condition, that reaches a patient and results in death, permanent harm or severe harm. Its policy, established in 1996, treats these as events that signal the need for immediate investigation and response.

The Joint Commission list also names events that are sentinel whatever their outcome. Examples include surgery or another invasive procedure at the wrong site, on the wrong patient or of the wrong kind, unintended retention of a foreign object after an invasive procedure, and death by self-inflicted injury within 7 days of discharge from inpatient care. Falls in staffed around-the-clock settings count when they cause injuries such as any fracture. The sentinel label therefore depends on the type of event as well as its severity. In the Joint Commission's 2024 data, 21% of reported sentinel events were associated with death and 49% with severe harm, while 21% involved moderate harm and 2% no harm.

The NHS in England uses a related but separate category. NHS England defines Never Events as patient safety events that are wholly preventable because strong systemic protective barriers exist at a national level. Sentinel events, Never Events and the wider class of adverse events overlap without being interchangeable, so a rate should name the list and the version behind its numerator.

Matching the denominator to each sentinel event type

A count becomes a rate when it is divided by a measure of how often the event could have happened. The general form is shown below.

$$R = \frac{N}{D} \times K$$

where $R$ is the sentinel event rate, $N$ is the number of confirmed qualifying events in the period, $D$ is the number of eligible opportunities or units of exposure in the same period and setting, and $K$ is a reporting multiplier such as 100,000 that makes small rates readable.

The denominator should count occasions on which the defined event could occur. A broad denominator such as all admissions dilutes a procedure-specific rate with patients who were never at risk. The categories in the Joint Commission list call for different denominators.

An aggregate rate that divides all categories by one denominator mixes unrelated exposure processes, so type-specific rates are easier to defend, with the overall count kept as a surveillance summary. Events also need a dating rule, by occurrence, discovery or confirmation.

Detection and reporting shape the numerator

The events counted are those that were detected, recognised as qualifying and reported. The dependence can be written simply.

$$R_{obs} = \lambda \times \pi$$

where $R_{obs}$ is the observed rate, $\lambda$ is the true underlying rate of qualifying events, and $\pi$ is the probability that a qualifying event is detected, classified correctly and reported.

Any change in $\pi$ moves the observed rate even when care is no safer. Because reporting of sentinel events to the Joint Commission is voluntary, the Commission states that no conclusions should be drawn about the actual relative frequency of events or trends over time, so its national counts divided by national activity would give a misleading incidence. The evidence on how much harm goes unreported is set out under adverse event reporting. A stronger reporting culture, encouraged by blame-free reporting, can raise the observed rate while true risk falls. Differences in $\pi$ between providers are rarely measured, so benchmarking is credible only when the event list, the denominator and the case-finding methods match, and a league table built on two or three events per provider can reward those that report least.

Small counts, exact intervals and years with no events

A Poisson distribution for the count is the usual starting model, and exact limits avoid the normal approximation, which can give negative lower limits for small counts. Ulm describes calculating exact Poisson limits through the link between the chi-squared and Poisson distributions.

$$L = \frac{\chi^2_{\alpha/2,\,2N}}{2}, \qquad U = \frac{\chi^2_{1-\alpha/2,\,2(N+1)}}{2}$$

where $L$ and $U$ are the lower and upper limits for the expected count, $N$ is the observed count, $\chi^2_{p,\,v}$ is the $p$ quantile of the chi-squared distribution with $v$ degrees of freedom, and $\alpha$ is 0.05 for a 95% confidence interval. The rate limits are $L$ and $U$ divided by $D$ and multiplied by $K$, and when $N$ is zero the lower limit is zero.

A year with no events does not show that the risk is zero. Eypasch and colleagues give the upper limit of the 95% confidence interval for an event not yet observed in $n$ patients as 3 divided by $n$, for $n$ above 30, the rule of three. The figure 3 rounds the one-sided 95% Poisson bound, the expected count at which a zero has a 5% chance.

$$U_0 = -\ln(0.05) = 2.996$$

where $U_0$ is the one-sided 95% upper limit for the expected count when no event has been observed.

Worked example of a surgical sentinel event rate over two years

All numbers in this example are illustrative. A surgical directorate performs 24,000 eligible invasive procedures in year 1 and confirms 2 events, one wrong-site procedure and one unintended retained foreign object. In year 2 it performs 25,000 procedures and confirms none. Case finding is assumed complete and unchanged.

1. Year 1 rate. The rate is 2 / 24,000 × 100,000 = 8.33 per 100,000 procedures.

2. Year 1 exact interval. For $N = 2$ the lower limit for the count is $\chi^2_{0.025,\,4}/2 = 0.484/2 = 0.242$ and the upper limit is $\chi^2_{0.975,\,6}/2 = 14.45/2 = 7.22$. Dividing by 24,000 and multiplying by 100,000 gives 1.01 to 30.1 per 100,000 procedures.

3. Year 2 upper limit. With no events the one-sided 95% upper limit is 2.996 / 25,000 × 100,000 = 11.98 per 100,000. The rule of three gives 3 / 25,000 × 100,000 = 12.0 per 100,000.

4. Two years pooled. Two events in 49,000 procedures give 2 / 49,000 × 100,000 = 4.08 per 100,000. The exact 95% interval runs between 0.242 / 49,000 × 100,000 = 0.49 and 7.22 / 49,000 × 100,000 = 14.7 per 100,000.

5. Interpretation. The fall from 8.33 to zero is well within chance, since with 2 events in total and unchanged risk the chance that both fall in year 1 is (24,000 / 49,000)² = 0.24. The pooled figure of about 4 per 100,000 is the better planning estimate, and the thirtyfold ratio between its limits shows how weakly two events pin down the true rate.

Monitoring rare events over time

Monthly rates of rare events are mostly zeros, which makes ordinary charts hard to read. Statistical process control separates the routine common cause variation of a stable system from signals worth investigating. For infrequent events Benneyan developed g and h control charts, which plot the number of cases between adverse events and can detect change with greater power than conventional binomial-based charts when event rates are low.

What the rate means for economic evaluation

The OECD report on the economics of patient safety estimated that 15% of hospital expenditure and activity in OECD countries can be attributed to treating safety failures. It found most of that burden associated with a few common adverse events, such as healthcare-associated infections and venous thromboembolism. Sentinel events are severe but rare, so a sentinel event rate captures a small share of that cost, and a cost-of-illness study of safety failures rests mainly on the common events. The same report notes that mandatory reporting of sentinel events was among the least favourably rated interventions in its expert survey.

When a rate does enter an economic model, the expected number of events avoided by an intervention follows directly.

$$E[\Delta N] = D(\lambda_0 - \lambda_1)$$

where $E[\Delta N]$ is the expected value of the number of events avoided per year, $D$ is the annual exposure, and $\lambda_0$ and $\lambda_1$ are the underlying rates per unit of exposure without and with the intervention. Costs and QALY losses are attached by event type, since a retained swab and a wrong-site amputation differ greatly in consequence.

Worked valuation of a surgical safety programme

All numbers in this example are illustrative. A safety programme costs GBP 50,000 a year in staff time and audit for 25,000 procedures. The baseline rate is 6 per 100,000 procedures and the programme halves it to 3 per 100,000. Each event avoided saves GBP 30,000 of treatment and avoids a loss of 2.0 QALYs.

1. Events avoided. $E[\Delta N] = 25{,}000 \times (6 - 3)/100{,}000 = 0.75$ events a year.

2. Costs and health gain. Savings are 0.75 × 30,000 = GBP 22,500, so the net cost is 50,000 − 22,500 = GBP 27,500. QALYs gained are 0.75 × 2.0 = 1.5 a year, so the incremental cost per QALY is 27,500 / 1.5 = GBP 18,333.

3. Net health benefit. The NICE manual (PMG36, section 6.3.2) asks for net health benefits presented at values of GBP 25,000 and GBP 35,000 per QALY gained, and the example borrows that convention.

$$NHB = \Delta Q - \frac{\Delta C}{k}$$

where $NHB$ is the net health benefit in QALYs, $\Delta Q$ is the QALY gain, $\Delta C$ is the net cost, and $k$ is the value placed on a QALY. At GBP 25,000 the net cost is worth 27,500 / 25,000 = 1.10 QALYs, so the result is 1.5 − 1.10 = 0.40 QALYs. At GBP 35,000 the net cost is worth 27,500 / 35,000 = 0.786 QALYs, so the result is 1.5 − 0.786 = 0.714 QALYs.

4. Sensitivity to the baseline rate. If the true baseline were 1 per 100,000 and the programme still halved it, events avoided would be 25,000 × 0.5 / 100,000 = 0.125. Savings would be GBP 3,750, the net cost GBP 46,250 and the QALY gain 0.25, so the net health benefit at GBP 25,000 would be 0.25 − 1.85 = −1.60 QALYs.

5. Interpretation. The decision turns on a baseline rate that a few events cannot estimate precisely, so pooled multi-year or multi-site data matter more than precise cost inputs. If the programme also reduces harms outside the sentinel list, counting sentinel events alone understates its value.

Sentinel event rate beside neighbouring safety measures

Several safety measures share the same arithmetic but count different events, and the choice depends on whether the purpose is learning from rare catastrophic failures or tracking the bulk of harm within quality of care monitoring. The sentinel event rate counts only events on a named list, many regardless of outcome. The adverse event rate counts adverse events more broadly, either as harm from care or, in trials, as events after treatment whatever their cause, the sense used as a model input. The harm rate tracks a defined adverse outcome per unit of person-time, and the near-miss rate counts events intercepted before they reach the patient.

Common misreadings of sentinel event rates

Most errors come from reading the rate as a direct measure of safety, detached from the definition and reporting system that produced it. Each can mislead about whether care has become safer or which provider is safest.

1. Reading voluntary national counts as incidence. The Joint Commission warns against conclusions about frequency or trends.
2. Treating fewer reports as safer care. Less reporting lowers the rate without any change in risk.
3. Treating a zero year as zero risk. The upper limit is about 3 divided by the exposure.
4. Waiting for a statistical signal. Each qualifying event calls for its own investigation, whatever the trend shows.

Sources

• Benneyan JC. Number-between g-type statistical quality control charts for monitoring adverse events. Health Care Management Science. 2001;4(4):305-318. https://doi.org/10.1023/a:1011846412909
• Eypasch E, Lefering R, Kum CK, Troidl H. Probability of adverse events that have not yet occurred: a statistical reminder. BMJ. 1995;311(7005):619-620. https://doi.org/10.1136/bmj.311.7005.619
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. Section 6.3.2. https://www.nice.org.uk/process/pmg36
• NHS England. Patient safety data. Statistics page, archived copy of 9 September 2026. https://web.archive.org/web/20260909202806/https://www.england.nhs.uk/statistics/statistical-work-areas/patient-safety-data/
• Slawomirski L, Auraaen A, Klazinga N. The economics of patient safety: strengthening a value-based approach to reducing patient harm at national level. OECD Health Working Papers No. 96. OECD; March 2017. https://www.oecd.org/social-issues-migration-health/the-economics-of-patient-safety_5a9858cd-en
• The Joint Commission. Sentinel Event Data 2024 Annual Review. 2025. https://digitalassets.jointcommission.org/api/public/content/eac7511986c0442a9c1ae04b1aa02cc0
• Ulm K. A simple method to calculate the confidence interval of a standardized mortality ratio (SMR). American Journal of Epidemiology. 1990;131(2):373-375. https://doi.org/10.1093/oxfordjournals.aje.a115507