Prevalence of Inpatient Pulse Oximetry in Operative and Nonoperative Settings

To the Editor:

Pulse oximetry’s earliest application was monitoring during surgery. Since the US Food and Drug Administration began regulating pulse oximeters in 1992, their use has proliferated outside the operating room without corresponding changes to regulatory guidance.¹ Pulse oximetry now influences clinical decision-making including triage,² interventions,³ and eligibility for respiratory equipment.⁴ Across these newer applications, arterial blood gas (ABG) confirmation or close observation may be less available than it is in the operating room to correct pulse oximetry errors, which may be common and disproportionately affect patients who are Black or Hispanic.⁵

Contemporary information about pulse oximetry use can help to ensure that pulse oximeters are tested in relevant clinical environments and inform policy and regulations, but is lacking. Therefore, in a multicenter sample of inpatients, we examined the frequencies of pulse oximeter monitoring in operative and nonoperative contexts and the observed distribution of pulse oximeter readings to establish the range over which performance testing is most relevant.

Methods

Database and Cohort

The Premier PINC AI Database (PINC)⁶ (patient admissions between January 1, 2016-September 30, 2022) is an enhanced claims-based multicenter database of approximately 25% of US hospitalizations with characteristics similar to those from the American Hospital Association Database.⁶ One-fifth of PINC hospitals also contribute vital sign information including pulse oximetry oxygen saturation (SpO₂) values.

We included patients admitted to hospitals that report SpO₂ data to PINC. Using room and board charges and dates of major surgeries,⁷ we categorized patient-days as occurring in a nontelemetry general ward, telemetry general ward, intermediate care unit, ICU, or operating room. For patient-days with > 1 care location, we selected the highest level of care. We excluded patients admitted to hospital care locations that had fewer than 10 patients per year with a recorded SpO₂ to increase certainty that all examined care locations reliably submitted pulse oximetry data to PINC.

Outcomes

The primary outcome was use of pulse oximetry monitoring on a patient-day. Among patient-days with pulse oximetry monitoring, secondary outcomes were SpO₂ and ABG use on a patient-day of pulse oximetry monitoring.

Statistical Analysis

We calculated percentages of patient-days with pulse oximetry overall and stratified by care location. We created a hierarchical logistic regression model (hospital random intercept) for pulse oximetry monitoring on a given patient-day (yes or no) with fixed effects for demographics, hospital characteristics, diagnoses,⁸ and care locations to quantify the median OR, a measure of residual heterogeneity on the OR scale attributable to admission hospital. For secondary outcomes, we reported the number and percentage of pulse oximetry-monitored patient-days with ABG use and the distribution of SpO₂ values stratified by care location and mechanical ventilation use.

Results

We identified 21,548,650 inpatient-days from 4,496,472 hospitalizations and 179 hospitals. Patients with included patient-days had median age of 64 years (interquartile range, 47-76 years). Patient-days were most frequently from hospitals in the South United Stated census region (59.1%) and with ≥ 500 beds (43.0%). Included patient-days most frequently were located in general wards (47.6%), followed by intermediate care units (25.1%), ICUs (14.9%), operating rooms (7.1%), and telemetry general wards (5.3%). Only 13.9% of patient-days had > 1 care location on a single calendar day; 99.9% of these patient-days included an operating room care location.

A total of 19,126,144 inpatient-days (88.8%) had pulse oximetry monitoring. The use of pulse oximetry was highest in intermediate care units (94.7% patient-days) (Fig 1A). Among all pulse oximetry-monitored patient-days, nearly one-half occurred on nontelemetry-monitored general wards (45.8% of pulse oximetry patient-days) and only 7.2% of all pulse oximetry-monitored patient-days occurred in the operating room. Moderate heterogeneity in pulse oximetry use was found among hospitals (median OR, 5.6 [95% CI, 5.5-5.7]) (Fig 1B).

Figure 1 –

A, B, Graphs showing inpatient-day pulse oximetry monitoring by care location. A, Bar graph showing the number of patient-days monitored by pulse oximetry by care location. The y-axis shows the count of patient-days and the x-axis shows care location. The height of the bar shows the total number of patient-days in that care location included in the cohort. The color of the bar denotes the number of patient-days with and without pulse oximetry use. B, Graph showing the predicted median hospital use of pulse oximetry by hospital. Shown are each hospital’s (x-axis) median (black dots) probability of pulse oximetry use and associated interquartile range (whiskers). Probabilities were calculated from the fitted hierarchical regression model by determining each patient-day’s predicted pulse oximetry use and then taking the median and interquartile range of these values per hospital.

ABGs were assessed during a total of 1,407,469 pulse-oximetry-monitored patient-days (7.4%), including in 155,364 operating room patient-days (11.3%). A total of 233,722,256 SpO₂ readings were recorded (approximately 12/patient-day). Patient-days in the operating room had numerically higher SpO₂ values compared with other care locations (Table 1).

TABLE 1 ].

Spo₂ Measurements by Care Location and Invasive Mechanical Ventilation Use

Care Location	Spo2, %
General ward
No mechanical ventilation	97 (95-99)
Mechanical ventilation	97 (94-99)
Telemetry general ward
No mechanical ventilation	96 (94-98)
Mechanical ventilation	98 (95-100)
Intermediate care unit
No mechanical ventilation	97 (95-99)
Mechanical ventilation	97 (95-100)
ICU
No mechanical ventilation	97 (95-99)
Mechanical ventilation	97 (95-100)
Operating room
No mechanical ventilation	98 (96-100)
Mechanical ventilation	99 (97-100)

Data are presented as median (interquartile range). Spo₂ = pulse oximetry oxygen saturation.

Discussion

We found that operating rooms now constitute a minority of the inpatient settings in which pulse oximeters are used. As such, those involved in testing and regulatory approval of pulse oximeters must consider that pulse oximeters need to perform accurately in settings where clinicians have little ancillary physiologic data and relatively brief interactions with patients. In these contexts, fewer opportunities are available to detect and correct for the falsely normal readings that have been shown to occur, to influence treatment, and to affect Black patients disproportionately.⁵ Because pulse oximeters are integrated into early warning, remote monitoring, and artificial intelligence settings, this limited clinical context for readings may become even more common.

So-called “indication creep” of approved medications and devices is well documented,⁹ but few examples in which treatment applications evolved from use in tightly monitored settings to near-ubiquitous use in settings with widely varying monitoring capabilities are available. Further devices are modified frequently, and new devices are approved using pathways for which few medication equivalents are available. The 510(k) pathway of “substantial equivalence” (a US Food and Drug Administration mechanism for candidate device approval based on comparisons with already approved devices or indications) therefore may be inappropriate when major changes in use context or technology occur that disconnect the way original devices were tested and approved from the way clinicians and purchasers need to evaluate contemporary devices.¹⁰ This may become more common with the advent of regulation of software as a medical device.

We were unable to differentiate between periodic and continuous monitoring. Although we limited our analysis to care locations with verified pulse oximetry data, a risk for false-negative results remains, especially at hospitals with near 0% predicted use. However, even if 100% of patients with ICU and OR patient-days received pulse oximetry, these locations would represent < 50% of all use. Assignment of patient-days with > 1 care location to the higher care level could overestimate use ICUs and ORs and could underestimate use in lower levels of care. We were unable to quantify oxygen delivery.

In conclusion, pulse oximeters have been disseminated widely in the inpatient setting. It may be important to align current approvals, particularly the US Food and Drug Administration’s 510(k) substantial equivalence pathway, with contemporary use cases.