Performance of a Blood Pressure Smartphone App in Pregnant WomenNovelty and Significance
The iPARR Trial (iPhone App Compared With Standard RR Measurement)
Hypertensive disorders are one of the leading causes of maternal death worldwide. Several smartphone apps claim to measure blood pressure (BP) using photoplethysmographic signals recorded by smartphone cameras. However, no single app has been validated for this use to date. We aimed to validate a new, promising smartphone algorithm. In this subgroup analysis of the iPARR trial (iPhone App Compared With Standard RR Measurement), we tested the Preventicus BP smartphone algorithm on 32 pregnant women. The trial was conducted based on the European Society of Hypertension International Protocol revision 2010 for validation of BP measuring devices in adults. Each individual received 7 sequential BP measurements starting with the reference device (Omron-HBP-1300) and followed by the smartphone measurement, resulting in 96 BP comparisons. Validation requirements of the European Society of Hypertension International Protocol revision 2010 were not fulfilled. Mean (±SD) systolic BP disagreement between the test and reference devices was 5.0 (±14.5) mm Hg. The number of absolute differences between test and reference device within 5, 10, and 15 mm Hg was 31, 53, and 64 of 96, respectively. A Bland–Altman plot showed an overestimation of smartphone-determined systolic BP in comparison with reference systolic BP in low range but an underestimation in medium-range BP. The Preventicus BP smartphone algorithm failed the accuracy criteria for estimating BP in pregnant women and was thus not commercialized. Pregnant women should be discouraged from using BP smartphone apps, unless there are algorithms specifically validated according to common protocols.
Arterial hypertension is one of the leading causes of maternal death worldwide. In developed regions, hypertensive disorders account for 14% of maternal mortality.1 Because of the high morbidity and mortality, specifically during pregnancy, it is of utmost importance to diagnose hypertensive disorders at an early stage to enable close surveillance and optimal treatment.2,3 Therefore, a reliable, readily accessible, and easily applicable blood pressure (BP) screening tool is desirable to detect hypertension at an early stage.
The use of smartphones and health applications (apps) is rising, especially among younger individuals. Several recent surveys showed that up to 64% of the German and US populations own a smartphone. Twenty-one percent to 58% use health apps regularly, converting a smartphone into a medical device. Among young individuals, aged 18 to 29 years, up to 85% possess a smartphone.4–6 There is a growing number of apps available in App Stores, which claim to measure BP using photoplethysmographic signals acquired by a smartphone camera. For smartphone users, this would be an inexpensive, convenient, and easily accessible option to self-screen BP, but compared with the large number (≈300 currently available)7,8 of validated automated devices for self-measurement of BP, no single BP monitoring app has been successfully validated to date.8,9
A new promising photoplethysmographic-based smartphone algorithm, estimating systolic BP (SBP), was recently tested on 1008 individuals at the University Hospital of Basel, Switzerland. The aim of this subgroup analysis was to evaluate this BP smartphone algorithm in pregnant women based on the European Society of Hypertension International Protocol revision 2010 (European Society of Hypertension [ESH]-IP 2010) for the validation of BP measuring devices in adults.10 Given that misdiagnosis of hypertensive disorders might lead to relevant health consequences for pregnant women and their unborn children, this subgroup analysis was done promptly after database closure, while the main analysis was still ongoing.
The data that support the findings of this study are available from the corresponding author on reasonable request.
The study protocol complies with the Declaration of Helsinki and was approved by the local ethics committee (EKNZ [Ethikkommission Nordwest- und Zentralschweiz] 2015–287). Before study inclusion, informed consent of subjects was obtained. The study was monitored by the Clinical Trial Unit Basel.
Device Details and Measurement Technique
The Preventicus BP smartphone algorithm (version 01/2017; Jena, Germany) estimates SBP based on photoplethysmographic signals recorded by a smartphone camera. For signal acquisition, the index finger is placed on the smartphone camera. Photoplethysmographic signals reflect the physiological blood flow in the vessels with each heartbeat. For further information on photoplethysmographic technique and the derivation of SBP, please see Methods in the online-only Data Supplement.
Study staff were familiarized with the predefined standard operating procedure (Table 1) for BP measurement in the iPARR trial (iPhone App Compared With Standard RR Measurement), and all 13 observers were trained by 1 ESH clinical hypertension specialist (T.B.) for the correct application. All subjects included in this subgroup analysis were recruited by the same observer. No problems were encountered during the test sessions.
The recruitment for the iPARR trial took place at the Department of Internal Medicine and the Department of Obstetrics and Gynecology at the University Hospital Basel, Switzerland (inpatient and outpatient clinics), between September 2015 and April 2016. Patients were included if they were aged ≥18 years, were able to give written informed consent, had a regular pulse, and did not have any contraindications for BP measurement (eg, dialysis shunt, lymphedema, and wounds). A total of 1008 subjects were included in the main study. Thirty-two of the 1008 subjects were pregnant and, therefore, included in this analysis (Figure 1). No subjects in this subgroup analysis had to be excluded because of arrhythmias, device failure, or poor quality of photoplethysmographic files.
The ESH-IP 2010 requests an entry BP (BPA) and a device detection BP (BPB) measurement. BPA is usually taken to ensure a uniform distribution of 33 subjects across the different predefined BP ranges, whereas BPB ensures that the test device is able to take a valid BP measurement.10 Because the recruitment of at least 1000 patients in the iPARR trial was planned, we expected to receive sufficient subject numbers across the different BP ranges and, therefore, omitted BPA. Because of this being a substudy of the iPARR trial, there was no choice for an even distribution of entry BP ranges in the cohort of pregnant women. Because all observers were blinded to SBP measured with the test device (Test-SBP) until the end of the recruitment phase, there was no need for BPB.
The validation procedure for the iPARR trial was developed based on the ESH-IP 2010 for the validation of BP measuring devices in adults.10 Because of feasibility regarding the large number of subjects planned to be included and concerns regarding mercury sphygmomanometers for clinical use in Switzerland, the following protocol modifications were necessary: the validated (mean error, −0.7 mm Hg; SD, ±4.0 mm Hg) automated upper arm oscillometric cuff device, Omron-HBP-1300, was used instead of mercury sphygmomanometers and binaural stethoscopes because this reference device could be used by a single observer.11,12 Hence, all SBP values measured with the reference device (Ref-SBP) and Test-SBP could be performed and recorded by the same observer. Test-SBP values were computed after recruitment closure and completion of all measurements by a technician who was blinded to Ref-SBP.
Measurements were performed in a comfortable surrounding without noise or other disturbances that could influence BP measurement. Subjects were sitting in upright position with legs uncrossed and back supported. The arm was equipped with an appropriate sized cuff (3 different sizes available) of the reference device and supported at heart level. After 5 minutes of quiet sitting, the first Ref-SBP was taken. After a 10-second pause, the smartphone measurement was started with the camera of an iPhone 4S (Apple, Inc, Cupertino, CA) placed on the patients’ index finger recording photoplethysmographic signals for 2 minutes. In total, each individual received 7 sequential BP measurements starting with the reference device and followed by the test device resulting in 4 Ref-SBP measurements and 3 Test-SBP measurements (Table 1). After recruitment was completed, all photoplethysmographic data files of the smartphone measurements were sent to Preventicus together with individual information on age, sex, height, and weight. Preventicus was blinded to Ref-SBP. Test-SBP results were returned to us to be merged with the corresponding Ref-SBP for final analysis according to the ESH-IP 2010.10 A total of 3 BP comparisons were obtained for each subject.
According to the ESH-IP 2010, differences between Test-SBP and previous Ref-SBP, as well as consecutive Ref-SBP, respectively, were calculated for each Test-SBP per subject. Then, absolute differences were determined. For BP comparison, the Ref-SBP that scored a smaller absolute difference from Test-SBP was used. If previous and consecutive Ref-SBP had the same absolute difference, the previous Ref-SBP was used. So called error groups—deviation between Test-SBP and corresponding Ref-SBP—were calculated, whereby a deviation of ≤5 mm Hg was termed as A, 6 to 10 mm Hg as B, 11 to 15 mm Hg as C, and >15 mm Hg as D, respectively. Based on these precalculations, Tables 2, 3, and 4 were generated to display validation results as proposed by the ESH-IP 2010 protocol.10 Predefined ESH-IP 2010 pass criteria were adapted for 96 pairs of comparison. For illustration of the results, Bland–Altman plots and Scatter plots were created. Spearman correlation coefficient between Ref-SBP and Test-SBP was calculated. Additionally, subanalysis was performed for low (Ref-SBP, <130 mm Hg) and medium (Ref-SBP, between 130 and 160 mm Hg) range. We calculated mean Δ, BP range, as well as Δ within ±5, within ±10, within ±15, and >±15 mm Hg, respectively, for the different BP ranges separately. All calculations were performed using IBM SPSS Statistics 22.
All 32 pregnant women were included for final analysis, resulting in 96 BP comparisons. Table 5 shows the characteristics of the substudy population. The mean (±SD) age was 31.6 (±5.1) years. Median (interquartile range) weeks of gestation were 37.5 (7.5) weeks. In 1 (3.1%) subject, preeclampsia was diagnosed at the time of recruitment. Table 2 shows the distribution of first SBP values measured with the reference device. Ref-SBP was ranging from 80 to 147 mm Hg. Eighty-two percent of all Ref-SBP were within low range, defined as SBP <130 mm Hg, whereas 18% were within medium range, defined as an SBP ≥130 and ≤160 mm Hg. No Ref-SBP was >160 mm Hg, which accounts for high range (Table 2). Mean (±SD) Test-SBP versus Ref-SBP disagreement was 5.0 (±14.5) mm Hg (Tables 3 and 4) and was larger when calculated separately for the different SBP ranges (Table S1 in the online-only Data Supplement). Bland–Altman plot showed an overestimation of Test-SBP in comparison with Ref-SBP in low range but underestimation in medium range (Figure 2). Spearman correlation coefficient was 0.401 (P<0.001) for all analyzed BP comparisons (Figure S2 in the online-only Data Supplement). Tables 3 and 4 display the validation results for SBP adjusted for 32 subjects with 96 BP comparisons. According to the ESH-IP 2010 criteria,10 the test device failed validation, achieving none of the validation requirements.
This study evaluated the accuracy of the Preventicus BP algorithm for SBP estimation in pregnant women. According to the ESH-IP 2010 criteria, adjusted for 32 subjects, there must be a minimum of 63, 79, and 90 of 96 BP comparisons within the deviation of ±5, ±10, and ±15 mm Hg, respectively, to pass validation. Further, at least 23 of 32 subjects must achieve at least 2 Test-SBP within a deviation of ±5 mm Hg from Ref-SBP, and ≤3 subjects can deviate in all Test-SBP >±5 mm Hg.10 Because the tested smartphone algorithm achieved none of these requirements, it is deemed to have failed the validation and was not commercialized. The British Hypertension Society protocol from 1993 differentiates between 4 grades (A, B, C, and D) of accuracy, where A is the most accurate, and D is the most imprecise.13 Corresponding to these guidelines for evaluation of BP measuring devices, the tested algorithm would be deemed to be category D—failure of validation.
This is the first study testing a BP smartphone algorithm on pregnant individuals. Currently, there are >500 BP apps available for download on Google Play Store and the Apple App Store. Most of them are declared as Prank Apps, but numerous ones claim to transform a smartphone into a medical device. Since 2014, the number of available smartphone apps measuring BP and pulse rate has continuously risen.9 Smartphone apps, which measure BP using photoplethysmographic signals, are highly popular, and some of them reached ≤1 million downloads (Google Play Store and Apple App Store; retrieved October 5, 2017; search terms: blood pressure and hypertension). However, there is still not a single validated BP app on the market.9 The last published validation study testing the top-selling Instant Blood Pressure app (AuraLife) on the general population did not reach accuracy criteria for BP measurement and was subsequently removed from App Stores.14,15 With the Preventicus app, we found the same pattern of overestimation and underestimation in the low and medium BP ranges, which could lead to overtreatment in women with normal values and undertreatment in women at risk for hypertension-related pregnancy complications—if these values were used for screening purpose or clinical decision-making. Given the results of both trials, the authors would like to state that all apps used for medical purposes should be tested in state-of-the-art clinical trials before being released in the market, and if they could not achieve the required accuracy, they should not be commercialized.
We, therefore, support the approach taken by Preventicus to conduct clinical validation trials and only release an app if clinical trial data provides evidence for its intended use. Vice versa, we question the accuracy of remaining photoplethysmographic-based BP apps.
Although the iPARR trial was designed based on the ESH-IP 201010, several adjustments had to be made because of the fact that we investigated a new technology for BP measurement and aimed to test it, not only for different BP ranges but also on a heterogeneous patient population during daily clinical routine. We, therefore, recruited a 30-fold larger number of subjects than recommended by the ESH-IP 201010. Thus, the procedure had to be adapted because of feasibility and recruitment strategy.
A major protocol deviation was the use of the automated upper arm oscillometric cuff device, Omron-HBP-1300, as a reference method instead of mercury sphygmomanometers and binaural stethoscopes, but in the recently published consensus document for a universal standard for the validation of BP measuring devices, the use of accurate automated, nonmercury devices is now permitted.16 The Omron-HBP-1300 passed validation according to the ESH-IP 2010 showing a mean (±SD) error of −0.7 (±4.0) mm Hg with a uniform distribution in Bland–Altman plot11 and passed American National Standards Institute/American Association for the Advancement of Medical Instrumentation/International Organization for Standardization 81060–2 revision 2013 criteria.12,17 The mean error of the Preventicus algorithm might have been smaller than the reported +5 mm Hg if we had used mercury sphygmomanometers. But regarding the rather high SD of 14.5 mm Hg, the clear overestimation of low BP values, as well as underestimation of medium BP values (Figure 2), and the clear failure to reach accuracy criteria, we assume that the Preventicus algorithm still would not have passed validation. Had this substudy passed validation, a further validation study with mercury sphygmomanometers and binaural stethoscopes would have been conducted but was now found to be unnecessary regarding the present results.
According to the ESH-IP 2010, additional BPA and BPB are recommended for categorization of subjects into different BP ranges to achieve an even distribution of BP values and to ensure that the test device takes valid measurements.10 As observers were blinded to Test-SBP, BPB was not applicable. BPA measurements were not part of the iPARR procedure because we expected to receive enough measurements for all different BP ranges when recruiting 1000 subjects. Given this recruitment strategy, there was no choice for even BP distribution, nor could the overall SBP range and maximum difference of SBP recruitment ranges be influenced (Table 2). The present substudy failed to achieve an even distribution within the different BP ranges, and calculations were performed on rather normotensive individuals. The lack of subjects in high BP range (SBP, >160 mm Hg) may, in part, be explained because high BP is usually seen as urgent clinical situation requiring pharmacological intervention in late pregnancy.
Combining the error of the Omron device and the error because of the uneven BP distribution and considering the fact that the Bland–Altman plot shows a progressive underestimation of BP with increasing SBP values (Figure 2), we postulate that the validation results and accuracy achievements would be even worse with an even distribution and less subjects in the low and more subjects in a high range. This is underlined by the fact that the mean (±SD) Δ is −14.2 (±8.4) mm Hg when being calculated separately for medium BP range (Table S1 in the online-only Data Supplement).
At least some subject requirements defined in the ESH-IP 2010 were not achieved. First, we recruited only 32 pregnant subjects, instead of 33 individuals and, therefore, adjusted the pass criteria to 32 subjects.10 As the present study is a subgroup analysis of the iPARR trial, we did not have a choice in the number of patients. But even if the 33rd subject would have been a perfect match between Ref-SBP and Test-SBP, the Preventicus BP smartphone algorithm would have failed validation. Pregnant women tend to be younger in age compared with the general population, and we, therefore, decided to reduce the age restriction from 25 years to 18 years.10 Because only 2 women were <25 years, this modification had no impact on the results.
Although we did not fulfill all requirements according to the ESH-IP 2010, our results emphasize that BP smartphone apps must be validated before clinical use. Additionally, there is a lack of validated BP monitors for use during pregnancy. The Medaval.ie webpage lists 2478 BP monitors, whereof 1965 lack validation information. Of 441 validated BP devices, only 33 BP monitors were validated in pregnancy, mainly belonging to the Omron, Microlife, and Nissei device families.7 Exploring the BP devices labeled as validated on the medaval.ie webpage,7 it is conspicuous that they are mostly validated according to the older and less stringent British Hypertension Society 1993 protocol, usually including 30 subjects,13 and that some devices passed in pregnancy but showed conflicting data in preeclampsia.18 According to a current systematic review by Bello et al,19 only 34% of validation studies in pregnant subjects are done without protocol violations. These findings emphasize that there is a need for strict validation of BP devices, especially for subgroups such as pregnant women.
Finally, the Preventicus BP algorithm was actually limited to SBP measurements. Even if the SBP measurement of this algorithm was valid, the lack of diastolic values would prevent this app from replacing validated oscillometric BP devices; however, it may have been used as a screening tool.
At this point of time, the Preventicus BP smartphone algorithm cannot be recommended for BP measurement in pregnant women. Both our results and findings from previously conducted trials in nonpregnant subjects emphasize that there is a need for validation and strict regulations for health apps. Although validation of the current BP smartphone algorithms failed, these findings should motivate, rather than discourage, developers of promising health apps to continue to work on their algorithms to improve them for future use. Nonetheless, photoplethysmographic-based BP smartphone apps may have a great potential to screen for gestational hypertension in pregnant women in the future. Because of the fact that nowadays smartphones are widely used, especially among younger individuals, a BP smartphone app would be an inexpensive and easily accessible way to self-monitor BP during pregnancy.
We would like to thank A. Winterhalder, Ulrike Schmitt, Cristina Granado, Doris Müller, and Craig Kingston for their support.
Sources of Funding
This work was supported by the Department of Clinical Research of the University of Basel and by Preventicus. All further expenses were funded by Jens Eckstein.
J. Eckstein holds 0.5% virtual shares of Preventicus. The other authors report no conflicts.
The online-only Data Supplement is available with this article at http://hyper.ahajournals.org/lookup/suppl/doi:10.1161/HYPERTENSIONAHA.117.10647/-/DC1.
- Received November 21, 2017.
- Revision received December 9, 2017.
- Accepted March 14, 2018.
- © 2018 American Heart Association, Inc.
- Bibbins-Domingo K,
- Grossman DC,
- Curry SJ,
- et al
- Ernsting C,
- Dombrowski SU,
- Oedekoven M,
- O Sullivan JL,
- Kanzler M,
- Kuhlmey A,
- Gellert P
- Smith A
- Krebs P,
- Duncan DT
- 7.↵Medaval. Blood Pressure Monitors. http://medaval.ie/device-category/blood-pressure-monitors. Accessed January 8, 2018.
- 8.↵Dabl Educational Trust. Blood Pressure Monitors – Validations, Papers and Reviews. http://www.dableducational.org. Accessed September 26, 2017.
- O’Brien E,
- Atkins N,
- Stergiou G,
- Karpettas N,
- Parati G,
- Asmar R,
- Imai Y,
- Wang J,
- Mengden T,
- Shennan A
- Takahashi H,
- Yokoi T,
- Yoshika M
- Meng L,
- Zhao D,
- Pan Y,
- Ding W,
- Wei Q,
- Li H,
- Gao P,
- Mi J
- O’Brien E,
- Petrie J,
- Littler WA,
- De Swiet M,
- Padfield PL,
- Altman DG,
- Bland M,
- Coats A,
- Atkins N
- Plante TB,
- Urrea B,
- MacFarlane ZT,
- Blumenthal RS,
- Miller ER III.,
- Appel LJ,
- Martin SS
- Husain I
- Stergiou GS,
- Alpert B,
- Mieke S,
- et al
- 17.↵Association for the Advancement of Medical Instrumentation; American National Standards Institute; International Organization for Standardization. AAMI/ANSI/ISO 81060–2:2013, Non-Invasive Sphygmomanometers - Part 2: Clinical Investigation of Automated Measurement Type. Arlington, VA: AAMI; 2013.
- Bello NA,
- Woolley JJ,
- Cleary KL,
- Falzon L,
- Alpert BS,
- Oparil S,
- Cutter G,
- Wapner R,
- Muntner P,
- Tita AT,
- Shimbo D
Novelty and Significance
What Is New?
This is the first trial testing a blood pressure (BP) smartphone algorithm using photoplethysmographic signals in pregnant women.
What Is Relevant?
The tested algorithm failed to reach accuracy criteria.
To date, all commercially available BP smartphone apps using photoplethysmographic signals are not validated.
Pregnant women should only use validated devices for BP measurement.
For now, pregnant women should be discouraged to use nonvalidated BP smartphone apps based on photoplethysmographic algorithms. Our results emphasize that all BP measurement devices must be validated before market release.