INTRODUCTION
One of the most common of all emergency department (ED) presentations is chest pain, raising the suspicion of acute coronary syndrome (ACS) [1]. The resources required for identifying myocardial infarction in this cohort are a major contributor to ED resource utilization and add to ED overcrowding. Although acute myocardial infarction (AMI) is diagnosed in only a small a minority of all suspected ACS patients (the US ED “rule in” rate is commonly <10%) [2–5], the high mortality and morbidity of AMI make early identification a high priority; it is only after AMI is ruled out that other less critical diagnoses can be considered. The current gold standard for AMI diagnosis is based upon clinical presentation, electrocardiogram (ECG) findings, and troponin testing [6,7]. While an ECG can be performed in minutes, serial troponin measurements may take hours to result [8–15]. This delay is further compounded when non-high-sensitivity troponin measurements are used.
Because of the large number of patients with possible ACS, the effect this potential diagnosis imparts on overall ED operations is large. Thus, even relatively small improvements in efficiency, e.g., replacing contemporary troponin assays with high-sensitivity platforms, ultimately results in a shortening of the mean length of stay for all ED patients [16]. The importance of shortening the ED length of stay cannot be overstated, as analyses of data from >15 million visits have demonstrated that longer ED length of stay is strongly associated with increased acute mortality for all patient groups [17–22]. Thus, expediting ED length of stay via more efficient evaluation of suspected ACS patients is an important goal in contemporary ED management.
A recently available, optical technology-based, noninvasive wearable device, called Infrasensor (RCE Technologies Inc), has the potential to significantly shorten the time required to evaluate suspected ACS patients. The Infrasensor is a wearable device (Fig. 1) that performs a rapid transcutaneous infrared interrogation of various biomarkers, including cardiac troponin I (cTnI). This device displays results after 3 minutes during which it acquires infrared (IR) molecular spectroscopic signals through the dermis, offering a bloodless and more accessible way to evaluate pathologic myocardial protein release [23]. The Infrasensor operates on the principle that different molecules absorb specific frequencies of IR light, thus creating unique spectral fingerprints. Employing the attenuated total reflectance method, IR light is directed into an internal reflection element, e.g., germanium or zinc sulfide. The light reflects internally within the internal reflection element and partially penetrates the skin, where it interacts with proteins of interest [24]. The reflected light carries the spectral information of specific proteins, which are then analyzed for time series associations with disease phenotypes of interest. More detailed descriptions of the genesis and evolution of the Infrasensor are provided elsewhere [25–28].
While the Infrasensor facilitates early chest pain assessment, there are few Infrasensor baseline assessments in the healthy US population. Thus, our purpose in this study was to collect baseline Infrasensor data in a representative sample of healthy individuals and compare the data to Infrasensor results obtained from a population known to have cTnI concentrations exceeding the 99th percentile of the cTnI assay to determine the discriminatory power of Infrasensor testing.
METHODS
Ethics statement
The study was approved by the local institutional review board at each of the 10 US hospitals (Chandler Regional Medical Center, Chandler, AZ, USA; Medical University of South Carolina, Charleston, SC, USA; The Ohio State University Columbus, OH, USA; Wake Forest University Health Sciences, Salem, NC, USA; University of Wisconsin, Madison, WI, USA; Oregon Health and Science University, Portland, OR, USA; University of Maryland, Baltimore, MD, USA; University of Missouri, Columbia, MO, USA; University of California, San Francisco, CA, USA; and Ben Taub, Baylor College of Medicine, Houston, TX, USA). Written informed consent was obtained from the patients prior to enrollment.
Study design and setting
After local institutional review board approval at each of 10 US hospitals, we performed a prospective evaluation of the Infrasensor device. After signed informed consent was obtained, a convenience sample of approximately 800 patients was enrolled and patients were assigned to one of two prespecified cohorts. The first cohort comprised healthy controls who were defined as subjects with normal vital signs and basal metabolic index who provided all “no” answers to the attached questionnaire (Supplementary Material 1) and thus were very unlikely to have elevated cTnI levels [29]. Healthy control subjects were recruited by age into three cohorts (<45, 45–60, >60 years) as well as by sex, race, and ethnicity, such that the final dataset was representative of the overall population of the United States.
If a patient met the healthy cohort inclusion criteria, a whole blood sample was obtained and processed to plasma within 1 hour, stored at –80 °C, and a simultaneous Infrasensor measurement was taken. Although research staff were unblinded to the Infrasensor measures, no Infrasensor cutpoints to provide guidance as to the clinical significance of any result have been defined. An assessment for high-sensitivity cTnI (hs-cTnI; Siemens Atellica IM, Siemens Medical Solutions) was subsequently performed at a remote central core lab with patients dichotomized based on an hs-cTnI that was elevated or not. Elevated hs-cTnI was defined as above the upper reference level published in the package insert. Subjects with elevated hs-cTnI were included in the analyses to determine if the Infrasensor, in a binary sense, would indicate if the hs-cTnI measurement was within the normal range.
The second cohort was a prespecified group of hospitalized patients known to have a cTnI measurement above the 99th percentile upper reference range level of the local standard of care cTnI assay, without any other inclusion or exclusion criteria so that any pathology with an elevated cTnI was enrolled. In these patients, Infrasensor measurements were taken at the same time as a single sample of blood was obtained. This blood sample was processed to plasma within 1 hour, and stored at –80 °C. Subsequently, hs-cTnI was measured in the same core lab that processed the healthy cohort samples.
The sample size for the healthy cohort was prespecified to be approximately 800 patients, and the known cTnI elevation cohort was prespecified to be approximately 100 patients. The sizes of these populations were approximated targets as exclusionary lab results (e.g., glycosylated hemoglobin >6.5%) were not known at the time of enrollment.
Statistical analysis
The entire cohort (Fig. 2) was statistically analyzed to extract population characteristics such as the means±standard deviations of normally distributed parameters and medians (interquartile ranges, IQRs) of non-normally distributed data. We then performed a binary classification-based logistic regression between core lab troponin test results and simultaneous Infrasensor measurements. Model performance of the Infrasensor, and its association with the laboratory-measured plasma troponin results using sex-specific cutpoints was assessed by calculating sensitivity, specificity, as well as positive and negative predictive values. A random subset, representing 20% of the total sample, was cordoned off and a machine learning model was trained on this group. We utilized an extreme gradient boosting (XGBoost) model [30] to analyze features extracted from optical sensor time series data. XGBoost is an implementation of the gradient boosting technique that combines multiple weak learners, typically decision trees, into a strong predictive model. It works by constructing an ensemble of decision trees sequentially, where each new tree is built to correct the errors made by the previous ones using a gradient descent algorithm that optimizes for both accuracy and speed. Age was used as an additional feature for the machine learning model along with the Infrasensor input. XGBoost does not require any preprocessing such as normalization since tree-based models are not sensitive to the scale of the input features. These models split data based on feature thresholds rather than calculated distances. The trained model was then tested on the excluded cohort data. Data are presented as C-statistics with 95% confidence intervals (CIs).
Infrasensor evaluation
Infrasensor performance was evaluated using a fivefold cross-validation method where the dataset was randomly partitioned into five nearly equal sized subsets. For each fold, one subset was reserved as the test set, while the remaining four subsets were combined to form the training set. The model was trained on the training set and evaluated on the test set. This cycle was repeated five times, with each of the five subsets serving as the test set exactly once and always containing 168 patients, with the healthy and elevated samples equally distributed across the five subsets. This method helps mitigate the variance associated with random selection of the train-test split, providing a more reliable assessment of the model's performance. Two diagnostic strategies were tested, consisting of “Infrasensor” only and “Infrasensor + age” parameters.
RESULTS
Overall, we enrolled 840 patients at 10 US hospitals from January 17, 2023, to November 1, 2023, of which 727 (86.5%) were assigned to the normal (nonelevated) troponin control cohort and the remaining 113 (13.5%) to the elevated troponin group. There were no adverse events that resulted from troponin measurements using the Infrasensor. Median age was 61 years (IQR, 52–71 years) for the elevated cohort and 48 years (IQR, 32–57 years) for the nonelevated. Overall, 416 (50.5%) were female, with 33 of 113 (29.2%) in the elevated cohort and 383 of 727 (52.7%) in the nonelevated cohort (Table 1). Patients with elevated troponin were more often male (70.8%), with a median age of 62 years (IQR, 53–70 years), and more likely to be Hispanic.
Combined with age, the optimal (defined as the highest number possible for each individual parameter) sensitivity, specificity, and negative and positive predictive values of the Infrasensor for identifying patients with an elevated troponin was 0.90, 0.70, 0.98, and 0.48, respectively, with a C-statistic of 0.90 (95% CI, 0.89–0.99) (Fig. 3). The plot indicating the C-statistic is a graphical representation of the performance of a binary classifier system as its discrimination threshold is varied and shows the trade-off between the true positive rate (sensitivity) and the false positive rate (1–specificity) for different diagnostic strategies. Overall, the C-statistic was 0.83 (95% CI, 0.67–0.99) for the Infrasensor alone, and 0.90 (95% CI, 0.89–0.99) for Infrasensor and age combined. The Infrasensor did not perform better when coupled with other variables.
The potential clinical value of the Infrasensor is demonstrated by the findings presented in Table 2. The diagnostic strategy employed was Infrasensor with age. Given a sensitivity of 98%, 328 (39.0%) of the total patient cohort would be diagnosed as having a negative cTnI (with a corresponding false negative rate of zero). Additionally, with a specificity of 99%, 66 patients (7.8%) would be diagnosed as having an elevated cTnI, of which 13 (1.5%) would be false positives. Ultimately, 394 (47%) of the total sample could have disposition decisions either assuredly ruled out or ruled in for ACS by a high specificity cutpoint within the 3-minute time requirement of the Infrasensor. Using a diagnostic strategy combining Infrasensor measurements and age provides a sensitivity of 98%. This would give 328 (39.0%) of the total patient cohort who would be diagnosed as having a negative cTnI (with a corresponding false negative rate of zero).
DISCUSSION
Our pilot data suggest that the Infrasensor has the potential to delineate patients with elevated versus normal cTnI within 3 minutes of use. This accuracy and speed are of critical importance for ED operations, and ultimately patient satisfaction. Furthermore, by optimizing the Infrasensor diagnostic threshold, this device yielded a sensitivity of 98% and specificity of 99%, along with a negative predictive value of 100%, highlighting its feasibility as an early assessment tool for ACS.
Rapid early diagnostic sensitivity is a historical holy grail in the evaluation of patients with suspected ACS. High-sensitivity plasma troponin assays have approximated this target, but at the expense of specificity. In addition, they suffer the limitations of requiring serial measures in high-risk patients and in those with early presentation, such that speed must then be traded to acquire accuracy. As a minimum, the threshold of a 99% sensitivity has historically represented the consensus target to exclude AMI in patients presenting with suspected ACS [31]. The specificity and sensitivity of the Infrasensor suggest that it may have a future role in the evaluation of these patients, particularly with model optimization using larger datasets of prospective ED patients with suspected ACS. Clinically, in the future, if a higher sensitivity and specificity combination is achieved, Infrasensor negative patients could be managed as if AMI has been ruled out, while Infrasensor positive patients would undergo additional evaluation identical to the current process for AMI diagnosis. Furthermore, the wearable status of the Infrasensor could allow its earlier use in environments beyond the ED, such as emergency medical services and ambulatory settings.
While in our pilot study range optimization provided a diagnostic answer in 47% of all patients in our sample, an undiagnosed gray zone of 53% of patients remained. This gray zone cohort will require future investigation to evaluate the potential of serial or continuous Infrasensor measurement to provide diagnostic accuracy in this cohort. However, as Infrasensor data acquisition is much faster than the existing standard of care, patients in the Infrasensor gray zone cohort would simply receive the existing standard of care, namely serial blood draw troponin measurements, without any significant time delays compared to current practices.
The ability to detect elevated cTnI within 3 minutes with accurate identification of all patients with acute myocardial injury (sensitivity of 100%) would have a profound impact on the contemporary evaluation of suspected ACS. The ability to identify AMI, obtained at the time of the ECG (rather than an hour or later when the lab-based tests results become available), would allow appropriate patient disposition and interventions at a much earlier time point than is currently available. Since ED length of stay is also the major driver of patient satisfaction, the ability to make immediate diagnostic decisions would be invaluable.
Additionally, the noninvasive nature of the Infrasensor will allow evaluations outside of hospital or clinic facilities. For example, the Infrasensor could provide a result before leaving the scene of an emergency medical services encounter. This would allow ambulance staff to more accurately make destination decisions, such as transportation of patients with elevated cTnI who could potentially benefit from cardiac catheterization to procedure-capable institutions rather than the nearest facility that may not have a catheterization laboratory.
Not unexpectedly in our study, patients with elevated hs-cTnI were older. The average age of elevated and nonelevated cTnI groups in our held-out test set was 60 and 45 years, respectively, while the total population had an average age of 47 years. This suggests that the higher performance of the model that included age as a moderator was largely due to the shift in the age balance between our elevated and nonelevated cohorts. Further model optimization in patients within acute care settings are needed to confirm the potential utility of the Infrasensor in early assessment of patients with chest pain.
It is important to note that the time a positive cTnI result is obtained is somewhat irrelevant as any elevated concentration is always associated with increased adverse events. However, a negative cTnI cannot be used to exclude AMI until at least 3 hours after symptoms have occurred [6]. This 3-hour time delay is to ensure that cTnI released from infarcted myocardium has a sufficient plasma concentration for assay detection. Measurements taken too early may result in false negative results, as cTnI released into the plasma has not yet risen to concentrations high enough for assay detection. Unfortunately, we were not able to record the timing of hs-cTnI measurements in our study, and thus the sensitivity of very early Infrasensor measurements could not be evaluated. However, even when used only in individuals who had symptom onset at least 3 hours prior to use, the Infrasensor may till serve as a preliminary screening tool to enhance patient triage.
Additionally, wearables can potentially be used for continuous cTnI monitoring, a feature that is not currently available for plasma. For high-risk patients, such a device could identify the potential of complications that would, at best, be diagnosed much later (e.g., during cardiac surgery when the patient is sedated). Finally, as a wearable technology, it could potentially be used at home as a preventive tool. By monitoring baseline levels over time, the Infrasensor could detect changes that could indicate an increased risk for cardiac events that would otherwise go undiagnosed (e.g., monitoring in patients with sustained tachycardia) until a catastrophic event.
Our study has several limitations. As an open-label, observational pilot evaluation with diagnoses independent of Infrasensor measurements, future investigations are needed where the ground truth is determined by a panel of adjudicators blinded to the transcutaneously obtained results after review of all the available clinical data. Additionally, we did not enroll all suspected ACS patients at all times after their arrival, nor did we evaluate the potential ability of continuous Infrasensor measurements to diagnose greater numbers of gray zone patients. Therefore, the utility of the Infrasensor must be evaluated in real life “all comer” populations to determine if it will work with varying levels of diaphoresis, hair, pigmentation, or tattoos for example, as well as in patients with very early presentations. These steps will help clarify the reasons for false positive Infrasensor results, and potentially allow their elimination. In addition, this study compared the Infrasensor findings with the Siemen's troponin assay; the Infrasensor findings may differ if compared to a different gold standard cTnI assay platform. Lastly, as the Infrasensor is not currently considered a standard part of care, no actual clinical decisions were made based on the Infrasensor results. Ultimately, a prospective trial, where clinical care is determined by Infrasensor results, will be required.
In conclusion, we found that the wearable transcutaneous Infrasensor identified patients with elevated cTnI with excellent specificity and sensitivity. However, future “all comer” prospective suspected ACS trials are needed before this technology is adopted into the standard of care.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation 
Supplement
Print 
