1. Introduction
Influenza, a contagious respiratory illness, poses a continuous severe health threat to people throughout the US and the world. In late 2017, the World Health Organization (WHO) estimated that a total of 250,000 to 500,000 annual deaths were associated with influenza infection based on data over 10 years ago [
1]. The annual number of flu-related deaths based on more recent data from a larger, more diverse group of countries increased to 650,000 [
1]. In the 2018-2019 influenza season, the Centers for Disease Control and Prevention (CDC) estimated that influenza infection was associated with over 35.5 million illnesses, over 16.5 million medical visits, 490,600 hospitalizations, and 34,200 deaths in the US [
2]. In the current pandemic of SARS-CoV-2 (COVID-19), the reported influenza activity in the US and globally is lower than expected, which may be impacted by the COVID-19 pandemic and needs to be interpreted with caution [
3,
4]. The Global Influenza Surveillance and Response System (GISRS) from the WHO suggests that the threat of influenza epidemics and pandemics persists during the COVID-19 pandemic and that countries are advised to remain vigilant and active in influenza sentinel surveillance when implementing COVID-19 surveillance [
5]. Given its high morbidity, influenza has also imposed significant healthcare costs and burdens [
6]. A previous study estimated that in 2003, the annual direct medical cost for influenza treatment was approximately
$10.4 billion in the US [
7], while the average annual cost for seasonal influenza in Italy in the period 1999–2008 was approximately US
$1.6 billion [
8].
The timely and accurate diagnosis of influenza infection is imperative so that antiviral therapy can be appropriately prescribed, unnecessary testing reduced, nosocomial transmission prevented, and thousands of hospitalizations prevented (especially among children and older adults). While vaccination helps to reduce influenza morbidity and mortality, the vaccine effectiveness varies from 11–69% year-to-year [
9], and influenza outbreaks can occur even in settings with 99% vaccination coverage [
10]. Even though the early antiviral treatment of influenza also reduces the probability of influenza-associated complications and mortality [
11], antiviral treatment is often infrequently prescribed in outpatient settings because of a lack of timely diagnostic testing which can be due to patients not seeking treatment on time as well as delays owing to testing procedures [
12]. Therefore, it is still necessary to deploy timely and accurate influenza diagnosis, and an improvement in diagnostic sensitivity would also improve influenza surveillance [
13].
Reverse transcription-polymerase chain reaction (RT-PCR) assays, which consistently demonstrate high diagnostic sensitivity, are viewed as one of the “gold standards” of influenza diagnostic methods [
14,
15,
16]. However, PCR is not appropriate for point-of-care (POC) deployment since it usually requires trained staff in laboratories equipped with specialized thermal cycling equipment and strict environmental conditions to prevent contamination [
17,
18,
19]. Although automated PCR systems are under rapid development which could reduce the turnaround time from hours to < 30 min, the issues of contamination, requirement of trained operators, and high cost of machine and test (about
$ 30 ~ >
$ 100 per test) still hinder their wide use in POC settings [
20,
21]. In contrast, rapid influenza diagnostic tests (RIDTs), which are antigen-antibody-based lateral flow immunoassays (LFAs), can be completed without skilled technologists in less than 30 min at a lower cost (about <
$15 per test and can be even cheaper), and their results can be observed visually in the POC setting [
14]. Further, RIDTs are approximately 20–50 times less expensive than PCR tests. As a result, RIDTs are the dominant method for screening influenza infections in POC settings.
However, the current RIDTs implemented in clinics suffer from a low analytical sensitivity, which results in many false negative diagnoses and thus a delay in antiviral treatment and an increase in the spread of the disease. The sensitivity of RIDTs varies between 10% and 70%, although the specificity of the tests is as high as 90% [
14]. The CDC also reported that many Food and Drug Administration (FDA)-cleared RIDTs suffer from low sensitivity in the detection of samples with low viral concentrations, thus demonstrating a low overall sensitivity (40%–69%) for all tested specimens [
22]. The detection sensitivities of RIDTs also vary by virus type [
23]. Even in outbreak settings, the overall sensitivity is not much improved, ranging from 58-79% for different influenza subtypes [
24].
Numerous efforts have been made to improve the sensitivity of RIDTs by developing novel LFA techniques for POC use. These efforts include assay kinetics optimization and signal amplification in test regions by chemical enhancement and reader use, such as through electrochemical, fluorescence, surface-enhanced Raman scattering, photothermal, and magnetic amplification [
25,
26]. Several orders of magnitude improvements in detection limits can be achieved with these novel techniques compared with traditional LFAs, as summarized in previous perspective papers [
26,
27].
The thermal contrast amplification (TCA) method was proposed as a photothermal amplification method to improve RIDT sensitivity. Compared with other signal amplification methods, TCA has the significant advantage of simple use. It can be used as a simple and direct add-on step after a commercial LFA without the need to modify or redesign any LFA components or reagents. In TCA, the specifically captured gold nanoparticle (GNP) labels in the test regions are excited by laser irradiation at their plasmon resonance wavelength. This excitation generates strong thermal signals that can be detected by IR sensors and quantified to represent the number of GNPs and, therefore, captured antigens. Our previous studies, as summarized in
Table 1, show that TCA can improve the LFAs’ analytical sensitivity by up to 32-fold for commercial LFAs [
28,
29], and even larger improvements (256-fold) can be achieved when TCA is implemented together with assay optimization and GNP design on LFAs [
30].
In real clinical POC use, however, more complicated reaction conditions are expected compared to those in standard antigen-dilution studies. In particular, patient samples can vary widely in viscosity, volume, and range of complex molecules, which may induce the non-specific binding of GNPs in the test region. These factors can impact the LFA performance and thus the TCA outcome. Therefore, a prospective cohort study is needed to evaluate the TCA-LFA diagnostic platform for POC use. Our previous preliminary cohort study [
31] reported that the TCA reader was able to identify ~50% of the false negatives from all 88 false negatives in clinical group A: Streptococcus RIDTs (QuickVue Dipstick Strep A Test, Quidel) against the PCR results. In this study, to further evaluate the TCA reader, a double-blind collaborative clinical cohort study was conducted on influenza A and B RIDTs (QuickVue Influenza A + B Test, Quidel) from a larger cohort of patients (
n = 345) with a local primary care clinic (HealthEast Grand Avenue Clinic, St. Paul, MN, USA). The personnel that operated the TCA reader on the clinical LFAs and confirmatory PCR tests were blind to each other to eliminate potential bias in TCA data analysis and thermal results interpretation. The results show that the TCA reader can substantially improve the sensitivity of the RIDTs (i.e., Quidel LFAs) by visual readout. The improvement in sensitivity achieved by the TCA reader in detecting influenza A was higher than that achieved in detecting influenza B. Although the specificity was compromised slightly by the TCA reader due to the nonspecific binding issues with the LFAs, the overall performance of TCA was still better than that of the visual readout of RIDTs based on comparison of their plots in the receiver operating characteristic space and F1 scores, which is a metric of the accuracy of the diagnostic method. It is also expected that the sensitivity of the TCA reader can be further improved by immediately reading the wet LFAs upon assay completion to eliminate the increase in noise that results from the drying of the LFAs.
3. Results and Discussion
The statistically analyzed TCA results from the 345 Quidel LFAs were compared with those from the visual readout and summarized in
Table 2. As can be seen, the sensitivity for both influenza A and B was substantially improved by TCA testing compared with visual readout (influenza A: from 0.32 to 0.49; influenza B: from 0.21 to 0.28). The sensitivity improvement achieved by TCA for influenza A is higher than that achieved for influenza B (relative increase of 53% and 33%, respectively), which is likely due to weaker binding of the GNP labels in the influenza B’s test line or the possibly lower influenza B viral load in the patient cohort compared to that of influenza A. Of note, the sensitivity and specificity of the visual readout of the Quidel LFAs in this cohort study are lower than the claimed values from the manufacturer’s trials (use an FDA-cleared influenza molecular assay as standard results). The claimed sensitivities are 0.815 and 0.809 for influenza A and B, respectively, whereas the specificities are 0.978 and 0.991, respectively [
39]. This discrepancy might be caused by a potential difference in manual sample collection and/or percentage of low viral load samples between our cohort study and that carried out by the manufacturer. A similar poor sensitivity with Quidel RIDTs was also reported in a previous cohort study in the 2000-2001 influenza season [
16]. Nevertheless, TCA could detect subvisual signals and help compensate for the poor sensitivity of the visual readout to a significant extent (relative improvement of 33–55%). The thermal signals of visual false negative samples are shown in
Figure 2. Approximately 25% of the visual false negatives for influenza A can be detected as true positive by TCA and approximately 9% for influenza B. This advantage of TCA to pick out visual false negatives is consistent with our preliminary cohort study for group A: Streptococcus diagnostics [
31]. Thus, the results validate the capability of TCA to detect subvisual, weak positives and improve the sensitivity of clinical LFAs. It is also noted that the cutoff value of thermal signals for influenza B in
Figure 2b is higher than that for influenza A in
Figure 2a. This is likely because the test line of influenza B is at the foremost position in the LFA facing the upcoming flow, thus leading to the maximal possible nonspecific binding of GNPs.
Even though the Quidel RIDTs had a low sensitivity by visual readout, they still had a high specificity (0.99) for both influenza A and B detection (
Table 2). Note that TCA can slightly lower the specificity compared with visual readout, as shown in
Table 2. However, the final specificities achieved with TCA are still high for influenza A and B detection (0.90 and 0.98, respectively) and comparable to the measured specificities (0.99) and those from the manufacturer’s cohort study (0.97–0.99) [
39]. The slight drop in specificity caused by TCA likely stems from the TCA reader amplifying the noise from nonspecifically captured labels at the test lines along with the signal from specifically captured ones. This hypothesis can be proven by inspecting the thermal signals from samples that were randomly selected from the 345 clinical LFAs in
Figure 3. This inspection shows clear overlap between the thermal signals from true positives and false negatives. Shifting the cutoff lines could cause either more false positives or more false negatives, which indicates that the limitation is caused by intrinsic nonspecific interactions within the LFA performance. It is also worthwhile to mention that background staining in LFAs can also adversely impact TCA performance. For example, when testing nasopharyngeal wash samples by BD Veritor
TM RIDTs (Becton, Dickinson and Company, Sparks, MD, USA), some of these RIDTs showed very strong background staining (see examples in
Figure S2) which substantially increased the uncertainty and variation of thermal results from subsequent TCA reading (data not shown here). Therefore, it is inferred that optimizing the assay itself, such as with an improved buffer kit, to reduce nonspecific interactions is critical for improving the specificity of the overall TCA LFA platform. This approach can also increase the signal-to-noise ratio, which in turn enhances the sensitivity [
34].
The overall performance of the TCA readout still surpasses that of the visual readout when comparing their statistical results (from
Table 2a–d) in the receiver operating characteristic (ROC) space in
Figure 4. Though an informal metric, one point in the ROC space is better than another point if it is closer to the upper left corner, i.e., coordinate (0, 1) [
40]. As such, the TCA results showed better performance than the visual results, as their plots are closer to the upper left corner and farther from the diagonal random guess line in
Figure 4. Furthermore, the F1 scores, i.e., the harmonic mean of the positive predictive value and true positive rate (
Table 2) to indicate a test’s accuracy, of the thermal tests for influenza A and B (0.55 and 0.36, respectively) are also higher than those of visual readout (0.48 and 0.32, respectively for influenza A and B), as shown in
Table 2. Although both the accuracy and F1 score are indicators of testing performance, the F1 score is more important in this study due to the imbalanced frequency (i.e., counting) distribution among true positives, true negatives, false positives, and false negatives (
Table 2) and the emphasis on false positives and false negatives. In short, the TCA results showed better performance than visual readings in the detection of influenza A and B when tested by Quidel QuickVue LFAs.
TCA readings on wet LFAs from the Quidel RIDTs exhibit better testing performance. The impact of testing a dry vs. a wet LFA on the TCA performance was examined with 34 patients’ nasopharyngeal wash samples, which were randomly chosen for another Quidel LFA test and subsequent TCA scan on both the wet and dry LFAs. The thermal results of the wet and dry samples are compared in
Figure 5. Out of the 34 samples, 4 true negative samples were used to determine the cutoff of the thermal signals (i.e., summation of the average thermal signal and 3 times the standard deviation). Due to the extremely imbalanced distribution of true positives, true negatives, false positives, and false negatives for influenza B, only the results for influenza A are presented in
Figure 5. The dry LFAs generally have higher thermal signals than the wet ones due to the smaller heat capacity of the membrane when devoid of liquid. However, the uncertainty in the thermal signals also increases, which substantially elevates the cutoff line in the dry results. As a result, more false negatives occurred in the dry results than in the wet results, although some reduction in the number of false positives also occurred. An overall statistical comparison is shown in
Table 3. The lower F1 score for the dry results indicates that the dry results are less accurate than the wet results. These results are consistent with our previous findings with group A: Streptococcus LFAs by TCA reading, in which wet LFAs exhibited a slightly higher sensitivity improvement and lower thermal noise than dry LFAs [
31]. Thus, we can expect TCA to give better performance when implemented on wet LFAs following assay completion.
The area of LFA diagnostics has been undergoing rapid change as both assay improvements and sample preamplification strategies along with reader systems are deployed. These approaches are leading to increased sensitivity, speed, and ease of use while maintaining low cost for application in POC diagnostics. [
26] Currently, our TCA team is working on reading algorithm improvement, miniaturization, and cost reduction of the TCA reader for eventual commercialization.