Choosing blood pressure thresholds to inform pregnancy care in the community: An analysis of cluster trials

To inform digital health design by evaluating diagnostic test properties of antenatal blood pressure (BP) outputs and levels to identify women at risk of adverse outcomes.


| I N TRODUC TION
Given the additional burden of maternal and perinatal morbidity and mortality associated with pregnancy hypertension, [1][2][3][4][5] the World Health Organization has listed the measurement of blood pressure (BP) as a marker of the quality of antenatal care contacts. 6 However, many automated BP devices are inaccurate in pregnancy, and entering observed values into clinical records is prone to errors, 7 particularly, the tendency for terminal digit preference, rounding values to '0' or '5'. 7 The 3AS1-2® BP measurement device (Microlife) was developed and validated in both normotensive and hypertensive pregnant women, according to the universal standards of the Advancement of Medical Instrumentation/European Society of Hypertension/International Organisation. 8,9 This has since evolved into the CRADLE Vital Signs Alert (VSA)® device that incorporates a 'traffic light' system to guide care by non-experts (e.g. community health workers). 10,11 Currently, the traffic lights are: (i) green when systolic BP (sBP) is <140 mmHg and/or diastolic BP [dBP] <90 mmHg; (ii) amber when either sBP is 140-159 mmHg or dBP is 90-109 mmHg; and (iii) red when sBP is ≥160 mmHg and/or dBP ≥110 mmHg.
Using population-level, 3AS1-2®-derived data from the three Community-Level Interventions for Pre-eclampsia (CLIP) trials, [12][13][14][15] we have shown that diagnostic BP thresholds in pregnancy should not be lowered to those suggested by the American College of Cardiology and the American Heart Association (ACC-AHA). 16,17 However, a recent Delphi consensus has supported using BP values that are 5 mmHg lower when measured in a community (such as at home), compared with office BP values; 18 this means that 140/90 mmHg in the office would be equivalent to 135/85 mmHg in the community, and 160/110 mmHg equivalent to 155/105 mmHg.
Using BP data from in-community measurement in the Community-Level Interventions in Pre-eclampsia (CLIP) trials, [12][13][14][15] our aim was to compare the relation to adverse maternal and perinatal outcomes for BP levels defined according to different outputs, i.e. sBP and/or dBP, sBP alone, dBP alone or mean arterial BP (MAP), and different levels.

| M ET HODS
This was a planned secondary analysis of data from the intervention clusters within the CLIP cluster randomised trials. These evaluated a community-level evidence-based care package for recognition of hypertension and the triage, initial treatment and transport of these women in intervention clusters in Pakistan (n = 10), India (n = 6) and Mozambique (n = 6) (NCT01192412), 2014-2017. [12][13][14][15] Trained community health workers performed antenatal (and postnatal) clinical assessment (including BP measurement using the 3AS1-2®-device, without traffic lights), guided by a digital health risk stratification app, PIERS-On-the-Move (Pre-eclampsia Integrated Estimate of Risk Score, POM); 19 risk stratification was based on the min-iPIERS model for women assessed at ≥20 weeks, and the POM app interface was co-designed with the community for digital health-supported antenatal care. Antenatal contacts were scheduled monthly from enrolment, and took place in women's homes. When assessment identified systolic hypertension (sBP ≥140 mmHg), the app guided initial treatment: (i) oral methyldopa 750 mg for sBP ≥160 mmHg; (ii) intramuscular magnesium sulphate 10 g if severe preeclampsia was suspected, based on at least one of: sBP ≥160 mm Hg, miniPIERS probability of adverse maternal outcome ≥25%, eclampsia or stroke; and referral of women to a comprehensive emergency obstetric care facility, within 4 or 24 hours, depending on the assessed risk. The POM app triggered according to systolic hypertension and other HDP-related triggers, but not according to dBP. 15 The

| BP and outcome data collection
Included in this analysis were all consenting pregnant women in the 22 CLIP intervention clusters, who had had received at least one standardised antenatal BP measurement by a trained community health worker, had delivered by trial end, and had completed follow-up. The in-community BP results were available on the mobile POM device. Hospitals did not have access to the 3AS1-2® monitor, as the focus was on community-level interventions, and hospital BP measurements were not included in the analysis. Postnatal community BP measurements were not included, as it is likely that the clinical outcomes being evaluated occurred prior to the postnatal period in the community. The 3AS1-2® monitor displays heart rate, sBP and dBP, but not MAP, which was derived using programmed algorithms within the device.
At each visit, an average of two BP readings were recorded, unless readings were ≥10 mmHg apart, in which case a third reading was taken and an average of the second and third readings taken as the BP in the POM app.
Trained surveillance teams collected outcome data through regular household surveys (except in India, where a prospective population-based surveillance system was established) until up to 42 days following the woman's delivery. All outcome data included in this study were adjudicated within the main trial by clinical adjudication committees, including at least one obstetrician, one paediatrician and one clinical trialist.
The highest antenatal BP values (based on the average BP during each visit) for each woman's pregnancy were extracted for analysis from the POM app. BP outputs included in the analyses are described in Table 1.

| Outcomes
Prespecified adverse outcomes (described in detail in Table S2) included: (i) a combined adverse maternal and perinatal outcome; (ii) a combined adverse maternal outcome; (iii) a combined maternal central nervous system (CNS)focused outcome (of maternal death, eclampsia, stroke or coma); (iv) perinatal death (stillbirth, early neonatal death or late neonatal death); (v) stillbirth. Each of the composite outcomes was calculated without double counting per pregnancy. When a pregnancy resulted in more than one component of a composite outcome, these were each counted as individual outcomes. In the case of multifetal pregnancies, one perinatal outcome was included in the composite per pregnancy (e.g. if a twin pregnancy suffered two stillbirths, only one stillbirth was counted in the perinatal composite).

| Statistical analyses
Descriptive statistics were used to summarise baseline maternal characteristics and maximal BP categories overall. We evaluated which BP output-based definition(s) and which BP level threshold(s) should inform traffic light algorithms, according to their associations and diagnostic test properties with adverse outcomes.
Associations of BP and adverse outcomes were evaluated using risk ratios (RR), estimated from Poisson generalised estimating equation and adjusted for age, education, parity and gestational age at enrolment. In the Poisson models, we assessed BP categories two ways; each BP category was mutually exclusive of others and compared with the normotension-1 group and each had a model fitted to estimate the RR of all categories at our above the category, compared with all categories below.
Test properties of BP category (for detection of the risk of adverse outcomes) were estimated by calculating sensitivity, specificity, positive likelihood ratios (+LRs) and negative LRs (−LR). As a screening test, we agree that BP measurement aims to be sensitive (≥80%), to identify as many women as possible who are at risk of adverse outcomes and to reassure those with a normal BP. As a diagnostic test, BP measurement aims to be specific (≥80%), to identify accurately those at increased risk and to minimise both unnecessary heath resource utilisation and negative impacts on the pregnancy experience from monitoring women who have normal outcomes. Positive LR was calculated as sensitivity/(1 − specificity), −LR as (1 − sensitivity)/specificity, and confidence intervals (CIs) were calculated using standard methods. 21,22 A +LR was considered to be fair (2.0-4.99), good (5.0-9.99) or excellent (≥10.0), and a −LR to be weak (0.21-0.50), good (0.11-0.20) or excellent (≤0.10). [21][22][23] In calculations of test properties, each BP category and those above it were compared with all those below it, to mimic how such a BP cut-off would be used in practice.
Statistical analyses were performed using R version 4.0.2. 24 The study is reported in accordance with STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines.  (Table 1). Most women were aged in their mid-20s, had less than a basic level of education and were parous. In general, women were enrolled in CLIP late in the first trimester or early in the second trimester, with most enrolling at less than 20 weeks' gestation. Almost 95% of women were normotensive (sBP <140 mmHg and dBP <90 mmHg) antenatally. Most women delivered at term, but one-quarter delivered preterm. Another quarter suffered a maternal, fetal or neonatal complication, which were mostly morbidity for the mother (about one in 10 women) or death of the fetus (42 per 1000 live births) or neonate (41 per 1000 live births).

| R E SU LTS
At 18 825 (13.5%) of 139 859 visits overall, the first two BP readings for either sBP or dBP differed by more than 10 mmHg, and a third reading was required. Rates were similar by country: 10 Table 2 shows that irrespective of the outputs used to define BP category, point estimates indicate a general dose-response association between higher BP category and greater aRR for adverse outcomes, for each category compared with normotension-1 (<135/85 mmHg). The risk usually more than doubled between normotension-2 and severe hypertension. Table 3 shows that irrespective of the outputs used to define BP category, when the diagnostic criteria for each BP category were used as a threshold for diagnosis of an abnormal BP, the risk of most adverse outcomes increased from normtension-2, again demonstrating a dose-response association.

| Diagnostic test properties
The event rates, sensitivity, specificity, +LR and −LR values for the relation between BP category and outcomes are presented according to BP outputs, as sBP and/or dBP (Table S3), sBP only (Table S4), dBP only (Table S5) and MAP (Table S6 and Figure 1).
None of the BP categories were able to identify women at low risk of developing adverse outcomes, as all sensitivities were <20% (and usually well below it) for all BP categories (however defined) and adverse outcomes. Negative LR values were at least 0.9 for all comparisons (Tables S2-S6).
Values of +LR varied by BP category and outcomes, with specificities of at least 89% (and usually over 95%) across the outputs used to categorise BP (Tables S2-S6). Table 4 summarises the +LR findings from Tables S2-S6. Point estimates indicate good to excellent diagnostic test performance (shown in dark amber) by non-severe hypertension-C and severe hypertension cut-offs, for the outcomes of maternal CNS composite and stillbirth, with severe hypertension showing good performance for perinatal death. For nonsevere hypertension-C, although the currently used sBP and/or dBP definition of each BP category showed only fair performance for these outcomes, the 95% CI were largely overlapping. Diagnostic performance was fair (shown in light yellow) and similar for the maternal CNS composite and stillbirth, between normotension-2 and non-severe hypertension-A.

| Main findings
In this planned secondary analysis of 21 069 pregnancies from three low-and middle-income country (LMIC)-based cluster randomised trials, higher BP categories were associated with increasing risk of maternal and perinatal adverse outcomes, irrespective of BP output (i.e. sBP and/dBP, sBP only, dBP only or MAP). Although no BP category could exclude risk for any outcome (with all −LRs at least 0.9), BP cut-offs of ≥160/110 mmHg, as well as ≥155/105 mmH, were good to excellent at identifying women at increased risk of adverse maternal central nervous system outcomes and stillbirth, with severe hypertension showing additional good performance for perinatal death.

| Strengths and limitations
This study has several strengths, including our large sample size, community-based recruitment of unselected pregnant south Asian and sub-Saharan African women, and the standardised BP measurements (albeit with serial, singlearm measurement, rather than a both-arm technique), 25 with a widely-used pregnancy-validated device. 8,9 These factors support the generalisability of our findings to the T A B L E 2 Adjusted risk ratios (95% CI) comparing BP categories with normotension-1 (sBP <135 mmHg and dBP <85 mmHg) and their association with CLIP outcomes, according to BP outputs. The CLIP composite outcome was a composite of maternal and perinatal mortality and morbidity. Maternal mortality or morbidity were assessed during or within 42 days of pregnancy; morbidity was defined as one or more life-threatening pregnancy complication, defined as a serious end-organ complication of pre-eclampsia, another major cause of maternal mortality, or receipt of a life-saving intervention. The maternal CNS composite was one or more of maternal eclampsia, stroke, coma or maternal death. Fetal/newborn death was stillbirth, early or late neonatal mortality. For outcome definitions, see Table S2.
development and validation of a digital health infrastructure applicable to pregnant women in similar LMIC settings. The POM app was co-designed with end users, 26 with the associated context-specific pictograms used in the app and for community engagement sessions co-designed with pregnant women, their families and maternity care providers. Societal and care provider barriers to, and facilitators of, the planned intervention were assessed using mixed methods. [27][28][29][30][31][32][33][34][35] The primary combined maternal and perinatal outcome for the CLIP trials were determined prior to the development of the pre-eclampsia core outcome set, iHOPE, 36 but represent reasonable LMIC community-level adaptations of those outcomes. There is no core outcome set for reporting broad maternity interventions at either community or primary health centre levels in LMICs.
This study has several limitations. First, as mentioned above, the 3AS1-2® monitor did not display the raw MAP values; these values had to be back-calculated from the displayed sBP and dBP, although MAP was actually measured directly, and the displayed sBP and dBP calculated.
T A B L E 3 Adjusted risk ratios (95% CI) comparing BP categories and those above it, with all BP categories below, and their association with CLIP outcomes, according to BP outputs. Note: BP was categorised as: normotension-1 (dBP <85 mmHg), normotension-2 (dBP 85-89 mmHg), non-severe hypertension-A (dBP 90-99 mmHg), non-severe hypertension-B (dBP 100-104 mmHg), non-severe hypertension-C (dBP 105-109 mmHg), and severe hypertension (dBP ≥110 mmHg). Risk ratios were adjusted for maternal age, maternal basic education (see Section 2), gestational age at enrolment and nulliparity. Risk ratios were estimated from Poisson GEE and adjusted for age, education, parity and gestational age at enrolment; for example, for non-severe hypertension-A, the comparison is for all women with sBP ≥140/90 mmHg versus all women with sBP <139 mmHg and dBP <89 mmHg. The exception to this approach was for the normotension-1 BP category, which was compared with all BP categories above it. Risk ratios were estimated from Poisson generalised estimation equation and adjusted for age, education, parity and gestational age at enrolment.
Abbreviations: BP, blood pressure; CLIP, Community-Level Interventions for Pre-eclampsia trials; CNS, central nervous system; dBP, diastolic blood pressure; MAP, mean arterial pressure; sBP, systolic blood pressure. a The CLIP composite outcome was a composite of maternal and perinatal mortality and morbidity. Maternal mortality or morbidity were assessed during or within 42 days of pregnancy; morbidity was defined as one or more life-threatening pregnancy complication, defined as a serious end-organ complication of pre-eclampsia, another major cause of maternal mortality, or receipt of a life-saving intervention. The maternal CNS composite was one or more of maternal eclampsia, stroke, coma or maternal death. Fetal/newborn death was stillbirth, early or late neonatal mortality. For outcome definitions, see Table S2.

F I G U R E 1
Positive likelihood ratios for BP outputs and thresholds for adverse maternal and perinatal outcomes. BP, blood pressure; CLIP, Community-Level Interventions for Pre-eclampsia trials; CNS, central nervous system; dBP, diastolic blood pressure; +LR, positive likelihood ratio; MAP, mean arterial pressure; N2, normotension-2 (sBP 135-139 mmHg or dBP 85-89 mmHg); NSH-A, non-severe hypertension-A (sBP 140-149 mmHg or dBP 90-99 mmHg); NSH-B, non-severe hypertension-B (sBP 150-154 mmHg or dBP 100-104 mmHg); NSH-C, non-severe hypertension-C (sBP 155-159 mmHg or dBP 105-109 mmHg); sBP, systolic blood pressure; SH, severe hypertension (sBP ≥160 mmHg or dBP ≥110 mmHg). Dotted vertical lines at 2 (weak diagnostic test performance), 5 (strong diagnostic test performance) and 10 (very strong diagnostic test performance). The CLIP composite outcome was a composite of maternal and perinatal mortality and morbidity. Maternal mortality or morbidity were assessed during or within 42 days of pregnancy; morbidity was defined as one or more life-threatening pregnancy complication, defined as a serious end-organ complication of pre-eclampsia, another major cause of maternal mortality, or receipt of a life-saving intervention. The maternal CNS composite was one or more of maternal eclampsia, stroke, coma or maternal death. Fetal/newborn death was stillbirth, early or late neonatal mortality. For outcome definitions, see Table S2.

BP categrory BP categrory
Secondly, it is possible that the diagnostic test performance of sBP in the CLIP trials was reduced through intervention, as the BP decision algorithm nodes of the POM app were triggered by sBP alone; 15,19 however, the associations with adverse outcomes of BP category defined by sBP, as well as the diagnostic test performance of BP category defined by sBP for identification of women at risk, were not inferior to when BP categories included or were restricted to use of dBP. Thirdly, many women did not have weekly BP measurements from 36 weeks' gestation to delivery, as specified in the CLIP protocol, so, as most pre-eclampsia arises at term, 1 we may have underestimated the diagnostic test performance of all BP categories. Fourthly, women with severe and non-severe hypertension were referred to facilities for antihypertensive treatment (about which we have no further information) and definitive care; such management probably attenuated the relation between higher BP categories and outcomes. Finally, these findings are derived from a single, The CLIP composite outcome was a composite of maternal and perinatal mortality and morbidity. Maternal mortality or morbidity were assessed during or within 42 days of pregnancy; morbidity was defined as one or more life-threatening pregnancy complication, defined as a serious end-organ complication of pre-eclampsia, another major cause of maternal mortality, or receipt of a life-saving intervention. The maternal CNS composite was one or more of maternal eclampsia, stroke, coma or maternal death. Fetal/newborn death was stillbirth, early or late neonatal mortality. For outcome definitions, see Table S2. albeit three-country, cohort and would benefit from confirmation in other LMICs.

| Interpretation in light of other evidence
These results are important to consider when determining triggers for clinical responses in digital health tools for maternity care (optimal prediction of adverse outcomes requires multivariable modelling that includes, but is not limited, to BP). First, MAP is the raw value measured by automated and semiautomated BP devices, with sBP and dBP derived from those values, using the associated pulse waveforms. Reassuringly, BP categories (defined using sBP, dBP, either or MAP) performed similarly in terms of diagnostic test performance. It is important to note that dBP tends to be more stable than sBP, the latter being more responsive to moment-to-moment stimuli, 37 and non-severe hypertension in young women tends to be isolated diastolic hypertension, as previously noted in this cohort. 38 Other than critical care specialists and anaesthetists, clinicians are most familiar with sBP and dBP, and only somewhat familiar with MAP, which was not superior.
Secondly, we believe that a safety element for the design of digital BP devices and related apps should include BP ≥155/105 mmHg as a 'red light' trigger when used incommunity (including home) BP monitoring and ≥160 mmHg in formal healthcare settings. This is consistent with the association of severe office hypertension and adverse maternal and perinatal events, similar to that for pre-eclampsia, 39 and the association of severe elevations in sBP with adverse outcomes in this cohort specifically. 38 This approach would mimic BP management used in the Control of Hypertension In Pregnancy Study (CHIPS) and Chronic Hypertension And Pregnancy trials; in CHIPS, the BP algorithm directed antihypertensive treatment to be offered at a dBP ≥90 mmHg, targeting a dBP of 85 mmHg, unless sBP reached 160 mmHg (CHIPS), irrespective of dBP; this algorithm optimised maternal outcomes without negative perinatal consequences. 40,41 Although the diagnostic test performance was only fair for the current hypertension trigger of 140/90 mmHg, the +LR values were similar for normotension-2 (i.e. BP 135-139/85-89 mmHg), suggesting that the latter is an appropriate target for action when BP is measured in the community. These data support the findings of the International Society for the Study of Hypertension in Pregnancy (ISSHP) Delphi consensus regarding home BP monitoring. 18 Importantly, no BP category, even normotension-2, had even a fair −LR, suggesting that a normal BP is not useful in identifying women at low risk of developing adverse outcomes in the future. In this dataset previously, we determined that the same was true for the 130/80 mmHg ('Stage 1 hypertension') and 120 mmHg systolic ('Elevated BP') recommended outside pregnancy by 2017 American guidelines. 16,17 A normal BP must be repeated on a regular basis to ensure that it remains normal, and should remain an auditable standard for antenatal care. 6 In this cohort of CLIP trial participants, we previously determined that infrequent digital health-supported community health worker CLIP antenatal contacts were associated with increased risks compared with women who received no CLIP trial contacts, whereas receipt of at least eight contacts (four in Pakistan) cost-effectively reduced the burdens of maternal, fetal and newborn deaths; 15,42 at least eight antenatal contacts per pregnancy is the standard suggested by the World Health Organization. 6

| CONCLUSIONS
The following conclusions and recommendations for digital BP device and related app design for the assessment of BP in pregnancy include: • 'Traffic light' thresholds for women measuring their BP at home should be 5 mmHg lower than when women are being assessed in formal healthcare settings. 18 This means that the 'amber' light would trigger at a BP of 135/85 mmHg, and the 'red' light at 155/105 mmHg, triggering appropriate care pathways. This could be done by addition of either a switch on the BP device or a toggle in an app that will set thresholds according to clinical setting (i.e. home versus formal ambulatory or inpatient healthcare setting). This should be addressed in future research. • Normal BP values are insufficiently reassuring in terms of estimating the risk of developing adverse outcomes in the future. • The care of pregnant women with hypertension is often provided by non-experts, and can be cost-effectively Likewise, the following conclusions and recommendations for digitally supported blood pressure measurement in pregnancy should include the following: • The relationships between blood pressure and adverse outcomes are largely equivalent when blood pressure is defined by systolic, diastolic, either or mean arterial pressure. • Our data suggest that digital blood pressure devices with embedded displays to guide clinical management (e.g. traffic lights) should be enabled to differentiate between community-based and facility-based measurements, perhaps with a toggle switch to identify a clinical setting. • Community-based thresholds should be set 5 mmHg lower than formal healthcare-based settings, meaning that an 'amber' light would be set at a BP of 135/85 mmHg, and a 'red' light at 155/105 mmHg. enabled by the provision of digital BP devices and evidence-informed decision support to reduce the burden of pregnancy hypertension-related life-threatening, lifealtering and life-ending complications. 15,42

AU T HOR C ON T R I BU T ION S
PD designed the concept of the CLIP trials and wrote the first draft of the manuscript. ZAB and LAM were the co-principal investigators of the overall CLIP trials project. MBB, ZAB, SSG, AAM, RNQ, KM, CS and ES were co-principal investigators of the individual CLIP trials. MV co-ordinated the CLIP trials. The PIERS on the Move App was co-designed by JMA, DTD, GAD, LAM and PD in collaboration with the country and University of British Columbia teams. The trial databases were designed and trial data managed by JL. The analysis plan was conceived by JNB, AS, LAM and PD. JNB and AS did the data analysis. JNB, AS and LAM accessed and verified the data. All authors had full access to all the data in the study, and read and approved the final version of the article. JNB had access to all data and had responsibility for its integrity and the data analysis. The corresponding author had final responsibility for the decision to submit the paper for publication.

AC K NO W L E D GE M E N T S
We acknowledge the Government of Mozambique, Province of Sindh and Government of India for their permission to integrate the three CLIP trials into their health systems, and the associated in-kind support. In particular, we thank the families of the 143 women, 2591 fetuses and 2677 neonates who died during the study period, who were willing to share their stories despite their grief.

F U N DI NG I N FOR M AT ION
Funded by the University of British Columbia (OPP1017337), a grantee of the Bill & Melinda Gates Foundation. The funder of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report.

C ON F L IC T OF I N T E R E S T S TAT E M E N T LAM and PvD report grants from the Bill & Melinda Gates
Foundation during the conduct of the study. All other authors declare no competing interests. Completed disclosure of interest forms are available to view online as supporting information.

DATA AVA I L A BI L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.