We studied 195 patients who exhibited normal blood pressures, i.e., <150 mmHg systolic or <90 mmHg diastolic regardless of any drug administration. Patients with peripheral arterial disease with an ankle brachial index below 0.9 were excluded because precise measurements of the tibial arterial pressure wave would have been difficult. Additionally, patients with heart rates higher than 100 bpm were excluded because the resultant wave inflections on APG signals would have been unclear. Finally, 168 patients were included in the study. Table 1 shows the clinical information.

Pulse volume waveforms at the brachium and ankle were recorded using the baPWV with a semiconductor pressure sensor (OMRON Co. Ltd., Tokyo, Japan). The subjects were examined while resting in the supine position. Fingertip PPG and baPWV were simultaneously used to assess arterial stiffness. The bilateral brachial and ankle blood pressures and arterial pulse waves were also simultaneously measured using the baPWV device. All waveforms were recorded for 20 to 40 seconds. The pulse wave velocity was calculated from the distance between two arterial recording sites divided by the transit time. The transit time was determined from the time delay between the proximal and distal 'foot' waveforms. The foot of the wave was identified as the commencement of the sharp systolic upstroke. The results obtained with the right-side and the left-side baPWV were almost identical (r = 0.9502). The mean baPWV of the right and left sides was reported in the present study.

PPG measurements were obtained from the index finger to analyze the vascular status while the baPWV was being measured. PPG recording was performed for a duration of 20 to 60 seconds, and the APG signal (the 2nd derivative of the PPG) was obtained at the same time. As shown in Figure 1, the APG waveform enables a more accurate recognition of the inflection points than the original PPG waveform, and it can show anacrotic or dicrotic notches. However, PPG signals are often distorted by poor peripheral perfusion or motion artifacts. A preamplifier was applied using a TAS9 VIEW (IEMBIO Co. Ltd., Chuncheon, Korea) device to locally digitize the signals, and the signals were subsequently filtered digitally using an FIR filter to remove motion artifacts. The APG features were thus automatically processed for statistical analyses. PPG signals, obtained with a preamplifier based on a microcontroller (MSP430F6638, Texas Instruments, Dallas, TX, USA), were band-pass filtered (0.6-15 Hz) to remove the effects of the respiratory rhythm and higher-frequency disturbances. Figure 2 shows the system configuration, including both peripheral circuits and embedded software. The 1st through 4th derivative algorithms of the PPG signal and its automatic detection algorithm for APG features were programmed into the microcontroller. The auto-gain control algorithm was embedded to adjust for sudden amplitude changes to obtain a limited peak-to-peak signal for the APG analysis. Although the PPG signal contains clinical information that can provide the vascular status, it is still difficult to process because it is sensitive to artifacts. For this reason, the PPG signal has not been widely used in healthcare devices or mobile systems beyond oximetry. The APG waveform can discern four systolic waves and one diastolic wave, namely, the a-wave (i.e., early systolic positive wave), b-wave (i.e., early systolic negative wave), c-wave (i.e., late systolic re-increasing wave), d-wave (i.e., late systolic re-decreasing wave), and e-wave (i.e., early diastolic positive wave) at the dicrotic notch (Figure 1). The e-wave location corresponds to the closure of the aortic valve and subsequent retrograde blood flow, and it can be used to monitor cardiac function [9]. The 4th differentiation of the PPG waveform in the software was programmed to detect the zero slope of the 3rd differentiated waveform to detect the peaks of the 2nd derivative (i.e., APG features). The signals were band-pass filtered in the peripheral circuit via the pulse width modulation (PWM) port of the microcontroller to improve the quality of the PPG signal. A low-pass 80-tap FIR digital filter to reduce noise and artifacts operates at the microcontroller clock frequency of 16 MHz to obtain the desired frequency response with the passband frequency and stopband frequency of 20 Hz and 30 Hz at -60 dB, respectively. Three direct memory access (DMA) channels for an FIR are applied for data transfer to proceed in parallel with minimum delay because the CPU operates other subroutine functions while the transfer is in progress compared to a conventional design without the use of DMA.

Statistical analyses were performed using MedCalc ver. 11.2 (MedCalc, Ostend, Belgium). Pearson correlation coefficients were calculated to evaluate the associations of the baPWV and the APG indices with age. A 95% confidence interval for the correlation coefficient (r) was calculated corresponding to the range of values that contains the true correlation coefficient with 95% probability. Multiple regression analyses were used to assess independent associations between one dependent variable and two or more independent variables. We evaluated whether a particular APG index was best associated with arterial stiffness, independent of other related indices. Statistical significance was defined as p < 0.05.