alternative location on the hand. (F) Representative ECG and PPG waveforms acquired in this manner from a healthy term neonate. (G) Comparison of vital signs calculated from the ECG EES and a gold standard. Temperature and PAT data are displayed without reference data because these measurements are only periodically acquired with conventional devices. RESEARCH | RESEARCH ARTICLE Downloaded from https://www.science.org on July 15, 2022 Healthcare) on healthy adult volunteers (n = 3) show excellent agreement for HR (mean difference = 0.1 bpm, SD = 2.55 bpm) and RR (mean difference = 0.3 bpm, SD = 0.95 bpm) as shown in Fig. 5, I and J, respectively. The PPG EES relies on similar NFC protocols, but with in-sensor analytic methods that not only reduce requirements on transmission bandwidth but also provide, when used in conjunction with adaptive circuits, crucial functionality for stable operation. Specifically, the processing in this case enables (i) dynamic baseline control to ensure that the input to the ADC on the NFC SoC lies within the linear response range and (ii) real-time calculation of SpO2 from the PPG traces (Fig. 6A). Here, the processing begins with application of a moving-average filter to the photodetector response from the red and IR LEDs. When the larger of these two averaged PPG amplitudes (typically that associated with the IR response) lies outside of a range that is optimal for the ADC (0.25 to 0.7 V), a programmable difference amplifier with voltage dividers at V+ dynamically adjusts the baseline level. The circuit shown in Fig. 6B demonstrates the operation where the governing equation is Vtr ¼ Rf Rs Vpre þ 1 þ Rf Rs Vþ ð1Þ where Vtr is the voltage output of the amplifier, Vpre is the voltage of the input signal, Rs is the input resistance, and Rf is the feedback resistance. The voltage divider at V+ with resistors Rd1 and Rd2 is governed by the following equation with Vref = 1.8 V: Vþ ¼ Rd2Vref R þ Rd2 þ Rd1 a0 16 þ a1 8 þ a2 4 þ a3 2 ð2Þ Sixteen different baseline states can be accessed via activation of binary values from four generalpurpose input-output pins (GPIOs; a0, a1, a2, a3) on the SoC (fig. S38), applied through an R-2R resistor ladder. Figure 6C shows dynamic control of the output voltage Vtr of a sinusoidal input signal (frequency = 50 mHz, amplitude = 40 mV, Voffset = –30 mV). Starting with the default setting of the GPIO ports (a0, a1, a2, a3; all high, or 1111), the baseline level automatically adjusts to lower levels associated as the value of Vtr drifts above the upper boundary of the specified voltage range, and vice versa as Vtr falls below the lower boundary. The result maintains Vtr in the allowed range. Figure 6D summarizes the operation in an actual PPG recording. Without this type of real-time, in-sensor processing (IR_Non in Fig. 6D), robust operation would be impossible: PPG signals would quickly drift outside of the narrow operating range of the ADC as a result of patient-to-patient variations in skin pigmentation and unavoidable, time-dependent fluctuations in optical scattering that result from Chung et al., Science 363, eaau0780 (2019) 1 March 2019 9 of 12 Fig. 8. Data collection from neonates in operating neonatal intensive care units. (A) Bland-Altman plot for HR using data from an ECG EES and a clinical standard. (B) Bland-Altman plot for RR using data from an ECG EES and a clinical standard. (C) Bland-Altman plot for SpO2 using data from a PPG EES and a clinical standard. (D) Representative results for PAT determined using combined data from an ECG EES and a PPG EES. (E to G) Differential temperature data collected from an ECG EES and a PPG EES for three recruited neonates with gestational ages of 28 weeks (E), 29 weeks (F), and 40 weeks (G). The other data presented here were collected from this same set of neonates. See fig. S40 for additional data. RESEARCH | RESEARCH ARTICLE Downloaded from https://www.science.org on July 15, 2022 micromotions relative to underlying blood vessels and subdermal structures (28). Calculating SpO2 involves determining the ratio (Roa) between the alternating and direct components of the PPG signals according to Roa ¼ ACRED=DCRED ACIR=DCIR ð3Þ for data from the red and IR LEDs (Fig. 6E). An empirical calibration formula determined by comparison to an FDA-cleared fingertip oximeter measurement (MightySat Fingertip Oximeter, Masimo) converts the Roa to SpO2 (Fig. 6F). Timedependent variations of SpO2 determined in this manner appear in Fig. 6G with demonstration in a decrease with a breath hold in an adult volunteer. The time-synchronized outputs from the ECG EES and the PPG EES allow for determination of advanced physiological parameters that are of high clinical value but not regularly collected in routine practice in NICUs. A key example is the measurement of pulse arrival time (PAT), defined by the time lapse between the maximum fiducial point in the ECG signal (R peak) and the corresponding minimal fiducial point in the PPG signal at valley as in Fig. 6H, as a direct correlate to systolic blood pressure (29, 30). Blood pressure is an essential physiological marker of perfusion, autonomic function, and vascular tone for critically ill newborns (31). Cuff-based blood pressure measurements with sphygmomanometers fail to provide continuous measurement, overestimate blood pressure in premature neonates (32), and pose a direct risk for pressure-related injuries (33). Although arterial lines offer a continuous measurement of blood pressure in neonates, these invasive interventions can cause thrombosis, hematomas, infection, and even death (34). Thus, the ability to capture PAT noninvasively and continuously would be of high clinical value in the NICU, with prior reports providing evidence that PAT correlates with blood pressure in infants (35–37). The Moens-Korteweg equation provides a linear relationship between PAT and BP (38, 39). Measurements of 1/PAT performed in processing of ECG and PPG data in the host (fig. S39), together with corresponding values of systolic BP captured using a sphygmomanometer on a healthy adult during a period of rest after exercising (running at 6 miles per hour for 15 min),