Monday, May 29, 2023

Accuracy of a smartphone application for blood pressure estimation in Bangladesh, South Africa, and Tanzania – npj Digital Medicine

Study design

This was an observational validation study in which we compared BP measurements taken by two independent trained nurses on a standard auscultatory cuff compared to the BP measurements taken by a smartphone application (a study participant placing his/her right index finger on the camera of a smartphone). For the purposes of the study, we used a research version of OptiBPTM called CamBP, which incorporates the same algorithms and functionality of OptiBPTM but blinds research nurses and research assistants to the BP values generated by the application. The assessment methods and study procedures were based on the guidance from the Consensus Statement published by the Association for the Advancement of Medical Instrumentation, the European Society of Hypertension, the International Organization for Standardization (AAMI/ESH/ISO)27,28 and ISO 81,060–2:2018 standards for validation of BP measurement devices27,28.

As this was the first time OptiBPTM was assessed in LMIC settings, we included a planning phase to prepare for the study. This included testing the electronic Case Report Form (e-CRF) and device configurations for CamBP, refining the manual of operations to standardize training and procedures across all sites, and applying collected data to train the CamBP algorithms. The planning phase resulted in an update to the CamBP algorithms, used as the final version for the assessment. In addition, an independent data review panel comprised of experts in BP validation, digital health, and obstetric care reviewed findings from the planning phase and provided recommendations that informed the validation study procedures.

The study was conducted in community settings of Gaibanda, Bangladesh and Ifakara, Tanzania for the general population, and Kalafong Provincial Tertiary Hospital outside Pretoria, South Africa for pregnant populations. Johns Hopkins University-JiVita Bangladesh led the Bangladesh site; Ifakara Health Institute led the Tanzanian site, and University of Pretoria Research Centre for Maternal, Fetal, Newborn and Child Health Care Strategies / South African Medical Research Council led the South African site. Data collection began in November 2021 across all three sites and was completed in mid-December 2021. Study teams conducted weekly monitoring of BP distributions to target the recruitment and achieve the ISO sampling requirements.


Eligibility was based on age 18–80 years and providing written informed consent; in addition, current pregnancy was a requirement for the enrollment in South Africa (age 18–50 years). Individuals were excluded if they had a BP difference between two arms >15 mmHg for systolic and >10 mmHg for diastolic as measured by cuff; an unstable cardiac condition or in need of oxygen therapy; were unable to place index or middle finger of right hand on smartphone camera; or if their resting BP was >160/110 for pregnant populations and 180/120 for general population. Individuals excluded for extreme BP were also referred for immediate care or managed within the hospital setting in South Africa. Bangladesh also conducted additional screening to exclude individuals with COVID-19 symptoms. Pregnant individuals were excluded from the general population recruitment in Tanzania and Bangladesh.

In Bangladesh, participants were recruited from a pool of community health workers (CHWs) and their spouses in the catchment area. To limit COVID-19 exposure, the study team contacted the CHWs to obtain preliminary consent over the phone, screened for COVID-19 symptoms and made an appointment for participants to come to the study center for potential enrollment and written consent. Towards the end of the study, the team employed snowball sampling methods to identify individuals for meeting the required BP distributions.

In Tanzania, participants were recruited from five wards in Ifakara town by asking individuals to come to a designated place (e.g., school, village government office) at a set date and time. Messages were communicated through radio, town meetings and local leaders. One week prior to the beginning of data collection, the Ifakara Health Institute team conducted a meeting with ward executive officers and co-developed an implementation plan together with the local leaders to determine identification of venues for data collection. Local leaders led the coordination and invitation of community members to selected venues for data collection on planned dates. Nurses screened individuals for exclusion criteria and informed consent. Additionally, to obtain the BP distributions, study teams identified participants from the Heart and Lung clinic in Ifakara for eligibility screening and possible enrollment into the study.

In South Africa, pregnant women across all trimesters attending ANC clinic at Kalafong hospital were recruited after being assessed by a physician for their routine ANC. Data collectors approached women until the target total of consenting and eligible women for that day was reached. Once the required sample for normotensive pregnant women was reached, the study recruited only women with hypertension. Pregnant women with hypertension were then assessed for pre-eclampsia. Recruitment of women with hypertension continued until the target sample of pre-eclamptic women was reached. To be more efficient in identifying potential pre-eclamptic participants, the study team also recruited from the antenatal ward.

Preliminary screening was conducted to ensure the individual had not participated in the planning phase, did not have an absence of fingers or limbs, or major heart conditions. Eligibility was established after informed consent and a more detailed screening of blood pressure, pulse, and oxygen saturation.

Sample size

Sample size calculations were based on recommendations from the Collaboration Statement published by the Association for the Advancement of Medical Instrumentation, the European Society of Hypertension, the International Organization for Standardization (AAMI/ESH/ISO)27,28. This statement suggests a sample size of at least 85 participants for general population, as well as sample size of 45 participants for pregnant women, if an independent general population 85-subject study has been completed successfully27,28. In accordance with these recommendations, the sample size was set to 100 general participants each for the analytical sample in both Tanzania and Bangladesh, and 60 pregnant participants for the analytical sample in South Africa. In addition, the analytical sample for the general population participants needed to demonstrate a minimum 30% participation rate of both sexes and the following BP distributions: at least 5% systolic ≤100 mmHg, 5% systolic ≥160 mmHg; at least 20% systolic ≥140 mmHg; at least 5% diastolic ≤60 mmHg; at least 5% diastolic ≥100 mmHg; and at least 20% diastolic ≥85 mmHg27,28. For pregnant populations, the analytical sample needed to be distributed along the following groups with a minimum of 20 participants in each group: (i) normotensive pregnant women <140/90 mmHg; (ii) hypertensive pregnant women without proteinuria >300 mg in 24 h and BP ≥140/90 mmHg; and (iii) pre-eclampsia, with proteinuria >300 mg in 24 h and DBP ≥90 mmHg27,28.


Each site’s data collection team comprised of one research assistant and two nurses. The teams received a 3-day training on the study procedures, which included components on the protocol, standardizing BP measurement, and study workflow. This training was conducted twice, prior to the planning phase and refresher training prior to this study. As part of the quality assurance and training to standardize BP measurement prior to data collection, each pair of nurse teams had to achieve inter-rater reliability of systolic BP (SBP) and diastolic BP(DBP) differences ≤5 mmHg for 45 out of 50 practice measurements and SBP and DBP differences between ≤10 mmHg for 48 out of 50 practice measurements.

The study used two data collection tools: an e-CRF installed on tablets and the CamBP application installed on Samsung S7 smartphones. The e-CRF was developed using the Open Data Kit (ODK) mobile data collection platform. The e-CRF collected eligibility requirements, informed consent, sex, demographic information, finger condition, medical history, and reference (cuff) BP measurements. The CamBP application collected information on the sex of the participant and an initial calibration value derived from the cuff. The CamBP application used this information to estimate BP values and generate the index test measurements. These BP measurements were recorded on the smartphone and transmitted to a server to produce a spreadsheet of the recorded measurements. A unique identifier was preassigned to all participants and used to link the data from the two data collection tools. The study team, software developers and participants were all blinded to the CamBP outputs at the time of data collection. The software developers also did not have access to the reference blood pressure readings to ensure independent analysis.

During data collection, reference measurements were based on the average of two nurses’ simultaneous BP measurements taken through a manual double stethoscope (BV Medical Teaching/Training Dual-Head Stethoscope). Nurses were blinded to each other’s measurement readings. A research assistant was tasked with recording the reference measurements into the e-CRF. Based on ISO 81060–2 criteria, each round of nurses’ values were considered as valid paired readings and averaged to derive the reference value for the round if their measurements did not differ by >4 mmHg27,28. The study required a minimum of four rounds with valid paired readings with inter-nurse agreement and successful signal capture from the application. In instances where the inter-nurse difference was >4 mmHg, the e-CRF calculations discarded measurements and the research assistant notified nurses to repeat rounds. Rounds could be repeated up to four times, with a total maximum of eight rounds per participant (Table 5)27,28. The readings taken of the participants represented the resting state (after at least 5 min of sitting) to fulfill the ISO requirements.

Table 5 Workflow for study participation.

Data management

From the full dataset with eligible participants and BP measurements, several exclusion criteria were applied based on ISO 81060–2 guidance (Fig. 2). Firstly, exclusions due to CamBP device failure (e.g., not capturing optical pulse waves within allotted time, poor quality of the signal, irregular heart rate, outlier) were dropped. Secondly, measurements with inter-nurse differences in BP readings of >4 mmHg were excluded. Thirdly, participants with less than three paired measurements were also excluded. Lastly, where the participant’s systolic BP differed by >12 mmHg or >8 mmHg in diastolic BP across 2 out of the 4 rounds of data collection, the study participant and all their respective measurements were excluded as per the Collaboration Statement27.

Statistical analysis

Univariable analysis was conducted to describe age groups, BP measurements, and sex. Analysis was conducted for each country separately. Age groups were categorized based on both a general (18–24, 25–29, 30–35, 36–49, 50–65, >65 years) and pregnant population distributions (18–24, 25– 29, 30–35, 36–49 years). For the pregnant population, the median and interquartile ranges of the hemoglobin values are described.

The analysis focused on measurements assessed against ISO 81060–2 criterion 1 and 227,28. For criterion 1, the mean error and standard deviation between the CamBP device value and the corresponding reference cuff value was estimated for each valid round of BP measurements. Results of the analysis were compared against the ISO 81060–2:2018 criterion 1 standard, in which the requirement for the device measurement to be validated by the cuff measurement is to observe a mean error of ≤±5 mmHg with standard deviation of the mean error of ≤ 8 mmHg27,28.

For criterion 2, each subject’s BP measurements were averaged. Subsequently, the standard deviation of the difference between the subject’s average BP measurement from CamBP and corresponding cuff values was derived. These values were evaluated against the maximum permissible standard deviation as function of the estimated mean error (mmHg) calculated from criterion 1 (Table 6)27,28.

Table 6 ISO Criterion 2—Maximum permissible standard deviation as function of the error (mmHg).

The protocol was approved by the WHO/HRP Research Review Panel. Ethical approval for both the planning phase and accuracy assessment and was obtained from the WHO Ethical Review Committee (Protocol A65932) as well as the following relevant national entities (Bangladesh Medical Research Council—(BMRC/NREC/2016–2019/07); University of Pretoria Faculty of Health Sciences ethics Research Ethics Committee (626/2018)—South Africa, NIMR/R.81/Vol.IX/3159—Tanzania).

All analyses were conducted in SAS version 9.4 (Copyright © 2016 by SAS Institute Inc., Cary, NC, USA) and R version 4.1.1 (R Core Team 2021. R Foundation for Statistical Computing, Vienna, Austria.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Source link

Related Articles

Leave a Reply

Stay Connected

- Advertisement -spot_img

Latest Articles

%d bloggers like this: