Statement of Ethics clearance
The study protocol submitted by Acculi Labs Pvt. Ltd., R. R. Nagar, Bangalore, Karnataka, India was approved by the Vagus Institutional Ethics Committee, Bengaluru, Malleswaram, Karnataka, India review board, which is registered with the Central Drugs Standard Control Organization, Ministry of Health and Family Welfare, Govt. of India (No. ECR/1181/Inst/KA/2019, dated 30-01-2020). [16]
Declaration of patients’ consent
Signed informed consents of all participants’ have been taken on the organization letterhead according to the declaration of Helsinki by the research team prior test. [17]
Study Design
Coding and computation
Computations are done using Python 3.9.8 (64-bits) on IDLE editor in Windows 10 OS.
Perimenopausal condition check
All web-enrolled subjects are checked for the signs symptoms of perimenopausal syndrome under the guidance of two trained gynecologists having an average experience of 20 years. Key signs and symptoms considered in this work are – (i) irregular menstruation (onset, duration, and flow), (ii) hot flush, and (iii) mood swings (stress, anger-anxiety, and depression or SAAD issues) as mentioned by Du et al. (2020) [18] Subjects with these conditions over the previous 6 months are recruited for the study.
Clinical Anger Scale (CAS)
CAS is a gold-standard screening and grading instrument, developed by Snell et al. (1995) which is reliable. [19] It has 21-items each with a four-point scale (0, 1, 2, 3). The sum of the scores gives the total score for the subject. Overall, there are four grades based on the CAS scores as (i) Minimal (score: 0–13), (ii) Mild (score: 14–19), (iii) Moderate (Score: 20–28), and (iv) Severe (Score: 29–63). [20] In this paper, CAS scores are the ‘dependent’ variables or responses, which are divided into four groups of CAS, mentioned below.
Lyfas
Lyfas is a novel smartphone-based, and non-invasive optical biomarker capturing tool. The tool has been developed using the principle of digital signal processing. [21] It can capture a total of 101 different digital biomarkers and is commercially available. These functional biomarkers are indicative of the psychophysiological state of an individual. [21] Lyfas works on two principles, Photoplethysmography (PPG) and Photochromatography (PCG). PPG measures blood volume changes in the microvasculature, while PCG measures the reflected light from various blood components such as cells and solutes. [21] The process is carried out using an optical sensor on the camera and its flashlight acting as an information capturing layer. [21] The next layer is a signal processing layer, which consists of the proprietary mathematical modeling and algorithms (a combination of heuristics-ML-AI), which converts the input signal into actionable metrics, which in turn captures the functional biomarker parameters system-wise [21]. These parameters were then validated in clinical settings (history, physical examination, and laboratory investigations) to detect several electromechanical and physiological activities, such as cardiovascular mechanics, hemodynamics, hemorheology, indicative hematology, and biochemistry in the test takers. [21] The study by Das and Chattopadhyay (2021), Lyfas has also been found reliable in predicting the cardiac risks in (i) Duchenne muscular dystrophy [22] and (ii) Chronic Obstructive Pulmonary Disease (Chattopadhyay and Das (2022)). [23] In another study by Chattopadhyay and Das (2022), Lyfas has also been found reliable in phenotyping the triad of hypertension-anger-anxiety in a vulnerable adult sample. [24]
The working principle of Lyfas has been elaborated step-by-step:
Step-1
Placing the index finger and lightly pressing on the rear main camera of the smartphone with the Android version 7 and above operating system, pre-loaded with Lyfas application
Step-2
Relaxed position with normal breathing and start the test after ticking the consent box and then follow the voice-guided steps of the test
Step-3
The camera light captures the capillary blood volume using the principle of Photoplethysmography (PPG), Arterial photoplethysmography (APPG), Photo chromatography (PCG), short Heart Rate Variability (120 seconds HRV), Mobile Spirometry (SPM), and Maneuvers like Orthostatic homeostasis to extract 101 clinically established digital biomarkers
Step-4
Grouping of biomarkers into various organ systems using its proprietary heuristics, ML, and finally AI algorithm
Step-5
AI-enabled analytics of these biomarkers to assess several psychophysiological states of the body and visualization, and finally
Step-6
Correlating analytics with clinical conditions.
Rationale of parameter-selection
HRV and its correlated optical biomarkers surrogate for cardiac autonomic modulation (CAM) to maintain psychophysiological homeostasis. Mental illnesses may cause cardiac autonomic neuropathy (CAN) by disturbing the sympathovagal balance. Therefore, the heightened sympathetic drive is reflected through high LF/HF and SD1/SD2, low pNN50, RMSDD, SDNN, which may be noted in the disorders with high sympathetic drive, e.g., generalized anxiety disorders, [25] schizophrenia, [26] obsessive-compulsive disorders, [27] bipolar disorders, [28] and in many other. A relatively elevated parasympathetic drive is evident in severe degree depression, where the sympathetic drive lowers but the parasympathetic drive remains unaltered. [29]
On the other hand, physical parameters, such as middle-Age, HR (increased vasomotor response to hormonal imbalance [30]), BMI (usually gained weight towards obesity [31]), systolic and diastolic BP (often elevates as essential hypertension settles in the body as age advances [32]), estradiol (which is a measure of the hypothalamus-pituitary axis, usually drops [33]), TSH (often is elevated to compensate for the increased demand of thyroxin in the body [34]), HbA1c (rises as type-2 diabetes settles down in the body [35]), and cortisol (a measure of the hypothalamus-pituitary-adrenal axis is often elevated due to the increment of stress [36]) have direct or indirect relationships with the perimenopause stage in the women’s life.
Construction of LASI
Optical biomarkers (SDNN, RMSSD, pNN50, SD1/SD2, and LF/HF), obtained by the Lyfas test and physical parameters (Age, HR, BMI, systolic and diastolic BP, estradiol, TSH, HbA1c, and cortisol), obtained by the laboratory test and physical examination are the ‘independent’ variables or factors. Statistically significant (p < 0.05, CI 95%) independent variables using Spearman’s correlation scores are used to construct Lyfas Anger Screening Instrument (LASI), which is then validated against CAS. It is important to mention here that, based on LASI parameters, the severity of anger has been graded/classified as ‘minimal’, ‘mild’, ‘moderate’, and ‘severe’ by a team of three psychiatrists (average experience of 10 years, each), based on their professional experience, who have no clue (i.e., blind) about the respective CAS scores. Table 4 shows the sample LASI data and the corresponding anger grade, which is then matched with that of the CAS scores to evaluate the efficiency of LASI in anger grading (see Table 5).
Pilot study
A total of 415 perimenopausal females are recruited through web invitations from the company website from October – to December 2021. Out of which 205 consists of the ‘case’ as they have a history of anger episodes as per the CAS. Among the cases, for the past six months, roughly about 36% have had angry outbursts (severe CAS), 24% have occasional but manageable anger episodes (moderate CAS), while the remaining 40% have ‘suppressed’ anger, i.e., they feel angry but never show up or express (moderate-to-severe CAS). The remaining 210 samples consist of the ‘healthy control’ group having minimal or mild anger episodes (minimal-to-mild CAS or nil anger). Informed consent has been obtained from each subject before the study. A recent (a month-old) laboratory data of HbA1c (to note the chances of insulin resistance, normal range < 5.7%), serum estradiol (E2, normal range 30–400 pg/ml), [37] TSH (normal range 0.35–4.94 mIU/L), [38] and cortisol (normal range 10–20 mcg/dl when taken between 6–8 am) [39] are collected. Heart rate (normal range 60–100 bpm), Age of the subject, BMI (normal range 18.4–24.9), systolic BP (SBP, normal value is less than or equals 110–120 mmHg), diastolic BP (DBP, normal value is less than or equals to 68–80 mmHg) levels are also noted at the time of the test. HRV biomarkers, such as SDNN (normal range 50–60 ms), RMSSD (normal range 60–80 ms), pNN50 (normal range 20–40%), SD1/SD2 (normal range 1–3), and LF/HF (normal range 1-1.8) are captured by Lyfas tests, [40] [41] [42] which are taken three times a day – 7 am, 2 pm, and 10 pm for the same period for examining the differences in readings at different times of a day. Anger episodes are noted daily as per the Clinical Anger Scale (CAS) [19] for the same two weeks period.
Statistics
z-score normalization
In this work, factor/parameter-wise z-score normalization is performed before conducting data analysis. Data normalization is an important pre-processing step to convert the values of the attributes within the same scale. The z-score method of normalization normalizes each value of the dataset in such a way that the mean of all values is ‘0’ and the standard deviation is ‘1’. [43] The advantage of z-score normalization is that it handles outliers better compared to the max-min normalization method. [44] Appendix-1 shows the data matrix following z-score normalization.
Data analysis
Descriptive data analysis (estimation of mean, median, max-value, min-value, and standard deviations) has been performed to note the spread of the original (i.e., non-normalized) data and can be seen in Table 1. [45] Kolmogorov-Smirnov tests, skewness, and kurtosis are performed to conduct the normality test [46] and data internal consistency or fidelity has been tested by computing Cronbach’s alpha (α) and can be seen in Table 2. [47] Later on, the strength of correlations (ρ, see Fig. 2) and the statistical significance (p < 0.05, CI 95%, see Table 3) of each LASI-factor/parameter with that of the CAS scores are obtained with the help of Spearman’s rank correlation (ρ). [48] Afterward, the strengths of agreements between each significant parameter of LASI and CAS scores are estimated using Bland Altman's reliability assessment (BARA) [49] presented in Fig. 3 (BARA). Finally, the efficacy (classification metrics) of LASI in terms of recall (R), specificity (SP), precision (P), accuracy (A), Youden’s index or j-stat (J), and fscores (F) are evaluated as the validation method (see Table 5). Figure 1 shows the flow diagram of the material and method applied in the study.