Development and preliminary validation of infrared spectroscopic device for transdermal assessment of elevated cardiac troponin.

Background: The levels of circulating troponin are principally required in addition to electrocardiograms for the effective diagnosis of acute coronary syndrome. Current standard-of-care troponin assays provide a snapshot or momentary view of the levels due to the requirement of a blood draw. This modality further restricts the number of measurements given the clinical context of the patient. In this communication, we present the development and early validation of non-invasive transdermal monitoring of cardiac troponin-I to detect its elevated state.
Methods: Our device relies on infrared spectroscopic detection of troponin-I through the dermis and is tested in stepwise laboratory, benchtop, and clinical studies. Patients were recruited with suspected acute coronary syndrome.
Results: We demonstrate a significant correlation (r = 0.7774, P < 0.001, n = 52 biologically independent samples) between optically-derived data and blood-based immunoassay measurements with and an area under receiver operator characteristics of 0.895, sensitivity of 96.3%, and specificity of 60% for predicting a clinically meaningful threshold for defining elevated Troponin I.
Conclusion: This preliminary work introduces the potential of a bloodless transdermal measurement of troponin-I based on molecular spectroscopy. Further, potential pitfalls associated with infrared spectroscopic mode of inquiry are outlined including requisite steps needed for improving the precision and overall diagnostic value of the device in future studies.

Introduction

Over 10 million patients present with chest pain[1] in emergency departments (ED) in the United States alone. Over 80% of these are due to non-cardiac causes, resulting in an unnecessary burden in the ED, revealing the need for an instant non-invasive screening technique that can streamline the ED workflows[1]. Furthermore, 1 out of 5 myocardial infarctions (MI) is asymptomatic (silent), leading to nearly 200,000 silent MIs each year in the US[2]. Therefore, the development of new technologies that can allow early non-invasive detection of myocardial injury is imperative.

Detection of cardiac troponins[3] to assess cardiac injury has been around since the 1990s. The state-of-the-art[4] troponin quantitation mechanism is based on immunoassays[5] involving the use of two or more antibodies, one of which is labeled, typically with a chemiluminescent tag, which adds another level of complexity[6] in the analysis. While immunoassays are highly developed and accurate, extensive sample preparation including cumbersome blood[7] draws are required. Furthermore, it mandates the logistics between physician and laboratory. Point-of-care (POC) assays are becoming increasingly available such as Abbott iSTAT[8] but suffer from low sensitivity making them incompatible with rapid rule-out algorithms[9]. High sensitivity immunoassays have advanced state of art possibilities for point of care and home healthcare modalities with innovations by startups like Luminostics[10]. Similarly, recent developments with microneedle patches[11] further demonstrate promise for longitudinal monitoring of the levels of inflammatory biomarkers. While the recent POC[12] solutions reduce time to test results, there still remains a dependency on blood draw coupled with lower analytical sensitivity compared to central laboratory testing, hence limiting their application toward effective discharge from the ED. Posited in this letter as a solution is infrared spectroscopy[13], a widely used and applied characterization technique due to its ability to probe into the material at the molecular level and hence an inherently sensitive mode of interrogation. The most appealing aspects include the benefit of minimal or no sample preparation required. There are innovative wearable devices based on functional-near-infrared light interrogation[14] such as, cortical hemodynamics[15], blood oxygenation[16], blood glucose monitoring[17], etc. However, these methodologies are not sensitive to the molecular composition of materials and therefore require labeling or tagging the molecule of interest for its detection. Molecular spectroscopy is one of the cornerstones of analytical tools used to study structural and compositional chemistry[18]. This technology typically involves the study of the interaction between mid-infrared (MIR) radiation with matter for which, as a diagnostic tool in a wearable format, there is no device yet to-date that has shown potential. There have been some endeavors leveraging the diagnostic capabilities of MIR with analytes such as in serum[19], urine[20], breath[21], skin[22] etc. Being well established as an indispensable tool in material science[23], infrared spectroscopy has been since applied in food, drug, environmental, forensics disciplines and importantly in the biomedical field such as cancer detection[24], and even cardiac care[25,26]. Another advantage[27] is that, testing using non-invasive devices like pulse oximeter, breathalyzers, and bilirubinometers, are exempt from CLIA regulations because a sample is not taken from the body. Recently, complex MIR spectrometers have been scaled down in footprint to fit into portable-sized benchtop devices such as the MZ5 by OceanInsight. However, there is no non-invasive wearable diagnostic device based on MIR spectroscopy. IR spectroscopy has been elusive[28,29] as an alternative to current POC solutions due to the following reasons[30]: (a) Signal to noise ratio strongly dictates the minimum detectable limit (b) Since all matters are sources of infrared radiation, efficiencies of IR-based devices can be confounded by stray light (c) The most sensitive mode of operation which is Fourier Transform IR spectroscopy requires a large footprint and is highly sensitive to mechanical vibrations, due to moving components thus often confined to ex vivo modalities. This correspondence outlines steps taken to circumvent or mitigate the mentioned challenges leading to an efficacious non-invasive device capable of risk-stratifying[31] ACS[32] patients based on Troponin-I (cTnI) levels.

Methods

The bloodless transdermal infrared spectroscopic device to assess elevated troponin-I was developed over four research phases.

Ex vivo exploratory research

First, an investigational study was conducted to determine the spectral features that are unique to cardiac markers such as cardiac Troponin I (Sigma-Aldrich, T9924-20UG), creatine kinase-MB (Sigma-Aldrich, C0984-100UG, and B-type natriuretic peptide (Sigma-Aldrich, B5900-.5MG), with the optical characterization of these substances in their pure form. A Nicolet ThermoFisher IS50 infrared spectrometer (IEN Labs, GaTech) employing a (single-bounce) diamond IRE, is used to identify a spectral signature for cardiac troponin. This allows one to optically detect and quantify the presence of that biomarker in a host substrate such as whole blood. De-identified healthy whole blood was procured and characterized to determine if there are any confounding overlaps in the absorbance peaks of blood and cardiac biomarkers.

Consequently, to confirm the efficacy of the ATR mode of interrogation, 30 biologically independent de-identified blood samples with the corresponding measurement of high sensitivity-cTnI values were procured in collaboration with the department of laboratory medicine at UCSF upon executing a Human Material Transfer Agreement. These blood samples that had already been collected prior to this study in accordance with applicable laws, regulations, patient consent forms and authorizations pursuant to Institutional Review Board. Further ethics approval and consent were not required as the samples were de-identified and anonymized in accordance with the Health Insurance Portability and Accountability Act. Blood samples included were those identified with a spectrum of Troponin-I values between the limit of detection (LOD) and upper reporting limit of the SOC assay namely, Advia Centaur[33] Siemens hs-cTnI. The blood samples were optically characterized ex vivo with the modality of total internal reflection using Nicolet ThermoFisher IS50 Fourier Transform infrared spectrometer (IEN Labs, GaTech) employing a (single-bounce) diamond IRE. Each blood sample to be characterized is one microliter in volume as deposited on the IRE. Each sample was measured in triplicates, with every repeat being an average of 32 co-added scans at a resolution of 4 cm−1.

Development of a transdermal kit

A benchtop Attenuated Total Reflectance (ATR) based spectrometer by OceanInsight was implemented in a Cardiac care setting in collaboration with West Virginia University, Heart & Vascular Institute. It was hypothesized that transdermal measurement in the indirect correlation with myocardial injury would provide insights toward possible troponin measurement capability. An optical measurement was performed on the thumb of 4 normal and 5 cardiac patients. A multi-variate cluster analysis was performed on this data employing Ward’s algorithm with squared Euclidean method focused on the wavelength range including some of the unique absorbance features relating to Cardiac Troponin-I with an interrogation window of 1.7 µm. The specificity of this window to cTnI as opposed to cTnT and other troponin isoforms is to be investigated in future studies. This unbiased heterogeneity analysis classifies data based on similarity.

Feasibility of point-of-care assessment

ATR-FTIR although proven efficacious, is challenging as a candidate for POC devices due to its large size and susceptibility to mechanical vibrations. Typically with FTIR, a wavelength sweep is done by moving the mirrors of an interferometer and consequently transforming the data from a spatial to frequency domain. However, there isn’t a need for a wavelength sweep as the wavelength ranges of interest, based on the ex vivo characterization of cardiac biomarkers, are previously determined. This allows for the selective sensitization of the infrared detector to the wavelengths corresponding to cardiac troponin-I by means of interferometric optical filters. As the next developmental step, an ambulatory non-invasive trans-dermal wearable device was deployed toward a pilot study conducted in the cardiac observation unit by recruiting patients with suspicion of ACS upon obtaining proper consent to obtain 24 biologically independent samples. The wrist wearable was installed on the patient’s wrist within 10 min of a SOC blood draw, with the cTnI levels reported using the CE certified Snibe Maglumi-1000 high sensitivity assay. This version of the optical sensor was designed to be portable (palm-sized) by using a broadband infrared light source, germanium IRE and, thermopile detector with specifically chosen filters sensitive to two optical ranges: the first is representative of the Amide II band which is used as an internal normalization reference, the second range such that cTnI would have the largest contribution to the absorbance. Due to the absence of moving parts and complex optical components, the small form factor was achieved while minimally affected by mechanical vibrations. The confounding effect of stray light was negated by pulsing the emitter at 4 Hz while polling the thermopile at 8 Hz thus recording both the on and off state of the emitter. A differential of these two states accounted for the extraneous light captured by the detector.

The pipeline used for the data analysis schema toward establishing a correlation between the proposed noninvasive device and standard-of-care data is illustrated in Fig. 1.

Fig. 1

Data Analysis Schema.

Schematic of the data analysis pipeline indicating the trajectory from raw optical data collected transdermally to actionable data used for correlation studies. Two channels refer to the intensities obtained from the two optical windows of the detector. Sample set refers to one complete 10 min data stream. (ATR = Attenuated Total Reflectance, SOC = Standard of Care, cTnI = Cardiac Troponin-I).

Clinical validation of a high-fidelity prototype

In two concurrent follow-on pilot studies, the performance of such a 4 channel sensor was evaluated on subjects recruited at the Zuckerberg San Francisco General (ZSFG) Hospital emergency department and Sengupta Hospital and Research Institute (SHRI), yielding 29 and 23 biologically independent samples respectively. These patients over the age of 18, were recruited agnostic of sex, ethnicity etc., strictly based only on the presentation of chest pain under suspicion of Acute Coronary Syndrome. The protocols were reviewed and approved by the corresponding Institutional Review Boards, and all participants signed a written consent form to participate. The patients were selected to represent a wide range of troponin values. Blood was collected in tubes containing lithium heparin, centrifuged at 1000 × g for 10 min. At ZSFG, the plasma was tested for cTnI using a high sensitivity assay on the Siemens Centaur Analyzer[34] (hs-cTnI, Siemens Healthineers). This assay has a LOD of 2.5 ng/L and a 99th percentile of 47 ng/L. In some cases, results from routine troponin testing were used and when that was not available, permission from the patient was granted to collect a fresh blood sample through an intravenous line or venipuncture. At SHRI, CE certified Snibe Maglumi-1000 high sensitivity cTnI assay was employed with an LOD of 0.01 ng/L and a 99th percentile of 19 ng/L. Within 20 min of blood collection for hs-cTnI testing, the 4 channel sensor wearable was installed on the underside of the patient’s wrist and left unattended for 5 min. Infrared readings along with accelerometer data were obtained on a continuous basis. When completed, the sensor was removed, and a file containing these readings were sent through a WiFi connection directly to RCE for data processing.

Statistical analyses

Categorical variables were presented as counts and percentages, and continuous variables as means and standard deviations. Receiver operator characteristics (ROC) plot displays performance of a binary classification method with discrete output. Elevated or non-elevated state of troponin was used as the binary output and optical device values as the input for the binary classification method. An R package[35] was used to obtain these ROC plots. In this library, trapezoids are used to compute the AUC indicating an average AUC of 0.853 for a bootstrapped model with sensitivity of 100% and specificity of 70.59%. Confidence intervals are computed with Delong’s method with bootstrap resampling[36]. Pearson’s correlation coefficient was used to indicate significant linear relationship among quantitative variables and regression analysis was done. A value <0.05 was considered as significant. All statistical tests were performed using Python, R and JMP.