Carle’s obstetric early warning score as a screening tool for critical care admission

Pregnancy is largely considered as a normal physiological process, but occasionally complications may lead to morbidity or even mortality. Despite advances in medical science, maternal mortality continues to be significant, especially in developing nations which have long-term interlinked, intergenerational and extensive impact both for the immediate family and the wider community1,2. It is seen that often deterioration in pregnant and postpartum women follows a continuum of events wherein a healthy pregnant woman can have severe morbidity which at times can lead to a near-miss situation and even maternal mortality3. Early warning systems (EWSs) are physiological ‘track-and-trigger’ systems which ensure timely detection and intervention and hence disrupt the chain of adverse outcome. Physiological changes in pregnancy warrant dedicated obstetric EWSs4.

The Confidential Enquiry into Maternal and Child Health (CEMACH) 2003-2005 report5 first introduced the use of the Modified Early Obstetric Warning System (MEOWS) in obstetric patients and continued to recommend its use in its further reports. Subsequently, the system was validated by several authors for different maternal outcomes6-10. However, CEMACH MEOWS was based on expert opinions without formal prediction model development methodology resulting in the assignment of arbitrary cut-offs to continuous clinical variables.

In 2013, Carle et al11 gave an obstetric early warning score system (Carle’s OEWS) for use in labour wards which was retrospectively validated in pregnant women using parameters of the first 24 h of intensive care unit (ICU) admission. Subsequently, the score was validated externally by Paternina-Caicedo et al12. Carle’s OEWS is one among the many OEWSs currently in use but stands out from the rest as it is statistically designed, clinically modified and internally validated. There continues to be the use of different regional versions of OEWS that vary in parameters and cut-offs13-17.

The use of OEWS appears promising for the early identification of a deteriorating pregnant woman, but more evidence and standardization are required18. The few studies that have been published were conducted retrospectively among individuals already admitted in critical care units and with limited study population8,12. Some studies have used maternal morbidity as the outcome measure which lacks a universal definition6,7,9,10,13. Maternal deterioration proceeds through a sequence of events as shown below3: healthy→morbidity→severe morbidity (requiring CCU admission)→near-miss→death. Taking morbidity as the primary outcome could pose a large burden on the already overloaded tertiary centres while taking mortality as the outcome measure would negate the very purpose of screening.

The present observational study aimed to prospectively evaluate Carle’s OEWS as a bedside screening tool among 1250 pregnant and postpartum women for the prediction of admission to the obstetric critical care unit (OCCU), which has been used as a surrogate for severe maternal morbidity.

Material & Methods

This prospective observational study was conducted in the department of Obstetrics and Gynaecology, Vardhman Mahavir Medical College & Safdarjung Hospital, New Delhi, over an 18 month period (November 2017 to April 2019) after procuring institutional ethical clearance. The procedures followed are in accordance with the ethical standards laid down by ICMR’s (Indian Council of Medical Research) Ethical Guidelines for Biomedical and Health Research on Human Participants (https://ethics.ncdirindia.org/asset/pdf/ICMR_National_Ethical_Guidelines.pdf).

Sample Size: The formula used for the sample size calculation is:

Where Z_α is value of Z at two-sided alpha error of five per cent and δ is 0.025

Taking expected area under the curve of modified obstetric early warning score (OEWS) for predicting high-dependency care unit (HDU) or ICU admission to be 0.6 with δ as 0.025, and five per cent level of significance, the calculated sample size was1230 individuals11. Hence, the total sample size taken was 1250.

Inclusion and exclusion criteria: One thousand two hundred and fifty obstetric women with a period of gestation ≥28 wk and admitted in labour wards of the study hospital and gave informed consent, were enrolled in this study. Pregnant participants requiring operative interventions or critical care at admission were excluded. However, stable participants enrolled in the study that later required operative interventions or critical care admission during the course of hospital stay were included.

Scoring: Carle’s OEWS was used in the study for the risk assessment of participants (Table I). The seven physiological parameters of Carle’s OEWS were documented first at admission by the treating physician: systolic blood pressure (SBP), diastolic blood pressure (DBP), respiratory rate (RR), heart rate (HR), per cent oxygen required to maintain SpO₂ ≥96 per cent (FiO₂), temperature (TEMP) and conscious level. These measurements were then repeated every four hours during the first stage of labour, every half hour during the second stage as well as for the first two hours after delivery (vaginal/operative) and then every 12 h till the individual was discharged or till she was admitted to OCCU according to the hospital protocol. The complete set of parameters was recorded at the defined intervals, and there were no missing entries.

The study participants were followed up for outcome measures. At discharge, the primary investigator (PI) allotted a score to the value of each parameter according to Carle’s OEWS, as given in Table I. The scores of each set of parameters were then added to reach an aggregate score. Recording of parameters by the treating physician and score calculation by the PI at the end ensured that there was no bias. It may be noted that this was purely an observational study and the management of the participants was according to the hospital protocol.

Study parameters: For evaluation, the participants were divided into two groups. Among the women requiring OCCU admission (Group 1), the highest aggregate score in the 12 h preceding OCCU admission was taken, and for the remaining who did not require OCCU admission (Group 2), the highest aggregate score of all scores was taken for plotting receiver operating characteristic (ROC) curves.

Admission to OCCU was taken as a surrogate marker for severe maternal morbidity and hence the primary outcome. The secondary outcomes included: (i) maternal near-miss as per the Ministry of Health and Family Welfare, Government of India guidelines19; and (ii) maternal mortality as defined by the World Health Organization International Classification of Diseases (WHO ICD) 201020. Further, the relative risk of individual OEWS parameters was calculated to predict OCCU admission.

All statistical analyses were carried out using the Statistical Package for the Social Sciences (SPSS) software version 21.0 (IBM SPSS Statistics for Windows, Armonk, NY, USA). The categorical variables were presented in number and percentage (%) and were evaluated using the Chi-square test. The continuous variables were presented as mean±standard deviation and median, and P value was computed using Student’s t test. P<0.05 was considered as significant. Area under ROC curve (AUROC) for Carle’s OEWS was determined to predict OCCU admission, near-miss and mortality. Diagnostic test was used to calculate sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).

Results

The mean age of women participating in this study was 25yr±4.11 (Table II). Majority of the study participants were primiparous (43%) and about half of them were booked (49.6%). There were 22.7 per cent referrals (Table II). A significant association between the booking status and OCCU admission was observed (P<0.001). Being a referral unit, the study population chiefly comprised women with one or more morbidities. Women with no morbidity were unlikely to have OCCU admission (P=0.003) (Table II).

Figure 1 presents the frequency of trigger of individual parameters of the OEWS chart in the study population. RR was the most commonly triggered parameter (47.5%). Most of the study participants (39%) were triggered to only score one. Whereas conscious level (0.5%) and fraction of inspired oxygen (FiO₂) required to maintain SpO₂ ≥96 per cent (1.2%) were rarely triggered but when these were triggered, then the highest score of three was reached, demanding immediate attention.

Close

Fig. 1

Frequency of individual parameters of OEWS chart in the study population.

Export to PPT

It was found that 76 (6%) of 1250 women required OCCU admission. Of these, 44 were also near-miss and there were six maternal deaths. One thousand two hundred and forty-four participants were discharged alive.

The most common condition requiring OCCU admission was eclampsia (15.8%), followed by sepsis (11.8%) and postpartum haemorrhage (10.5%) (Table III).

The AUROC curve of OEWS for predicting OCCU admission was 0.975 with a sensitivity of 96.05 per cent, specificity of 89.1 per cent, PPV of 36.3 per cent and NPV of 99.7 per cent at a cut-off of 4.5 (Fig. 2 and Supplementary Table I).

Close

Fig. 2

ROC curve showing sensitivity and specificity of OEWS in predicting critical care admission. ROC; receiver operating characteristic.

Export to PPT

The AUROC curve of OEWS for predicting a maternal near-miss event was 0.971 with a sensitivity of 95.5 per cent, specificity of 94.7 per cent, PPV of 40 per cent and NPV of 99.8 per cent at a cut-off of 5.5 (Fig. 3 and Supplementary Table II).

Close

Fig. 3

ROC curve showing sensitivity and specificity of OEWS in predicting maternal near-miss.

Export to PPT

The AUROC curve of OEWS for predicting maternal mortality was 0.996 with a sensitivity of 100 per cent, specificity of 97.9 per cent, PPV of 23.5 per cent and NPV of 100 per cent at a cut-off of 8.5 (Fig. 4 and Supplementary Table III).

Close

Fig. 4

ROC curve showing sensitivity and specificity of OEWS in predicting maternal mortality.

Export to PPT

Overall, it was found that an aggregate score of zero defined a normal category, an aggregate score of ≤4.5 suggests low-risk individuals, while at an aggregate score of ≥4.5, obstetric cases needed CCU admission, at ≥5.5, one can end up as near-miss and mortality was high among individuals with a score of ≥8.5.

Figure 5 summarizes the relative risk of OCCU admission associated with each of the triggered OEWS parameters. Maximum relative risk was for RR (41.12) followed by per cent O₂ required to maintain SpO2 ≥96 per cent (20.25) and consciousness level (17.77). DBP was the only parameter found to be non-significant (P=0.12).

Close

Fig. 5

Relative risk of individual parameters in predicting critical care admission.

Export to PPT

Discussion

In the present study, no significant association was found between OCCU admission and maternal age or parity. This was in concordance with the observations of Paternina-Caicedo et al12. This only reiterates that all pregnancies have the potential to become high risk during labour and highlights the need for a universal safety net.

Seventy per cent of the individuals requiring OCCU admission received no previous antenatal care. Singh et al7 also found that unbooked participants were more likely to have morbidities.

It was observed that RR (47.5%) and SBP (32.2%) remained the most frequently triggered parameters. This was in contrast to all other studies validating CEMACH MEOWS6,7,9. This can be explained by the fact that CEMACH MEOWS uses a higher trigger of 21/min for RR and 151 mmHg for SBP, while in Carle’s OEWS, RR and SBP trigger at 18/min and 140 mmHg, respectively. DBP triggered at a comparable frequency in all studies6,7,9. Temperature triggered in 7.8 per cent of our population which may be explained by the fact that 13 per cent of women in our study had presented with prolonged pre-labour rupture of membranes.

The AUROC curve of OEWS for predicting OCCU admission was 0.97 with a sensitivity of 96 per cent, specificity of 89.1 per cent, PPV of 36.3 per cent and NPV of 99.7 per cent at a cut-off of 4.5. At an aggregate score of 1.5, the sensitivity was 98.7 per cent and specificity was only 33.6 per cent but improved at 4.5 aggregate score (Supplementary Table I). This justifies the concept for an aggregate score-based warning system over the mere absence or presence of a trigger.

Ryan et al8 observed that the specificity of CEMACH MEOWS for predicting OCCU admission improved from 54 to 72 per cent when only ≥1 red trigger is taken as a criterion than when ≥1 red or ≥2 amber triggers are considered, but this continues to be much less than the specificity of 89.1 per cent of Carle’s OEWS at a cut-off score of 4.5 as was seen in our study. Further, high NPV (99.7%) ensured that no deteriorating patient was missed. Low PPV (36.3%) warrants its utility as a screening tool, and not a diagnostic tool.

It was observed that Carle’s OEWS was more predictive of maternal near-miss at an aggregate score of 5.5 with a sensitivity of 95.5 per cent and specificity of 94.7 per cent as compared to another study in the Indian context which used CEMACH MEOWS and had a sensitivity of 70 per cent and specificity of 84.4 per cent9.

Further, the Carle’s OEWS was found to be useful in predicting maternal mortality with AUROC curve of 0.996 (CI: 0.092-1) (P<0.001), which was also observed by others11,12. These studies, however, did not define the cut-off used and were conducted in ICU patients. We observed that at an aggregate of 8.5, Carle’s score was 100 per cent sensitive and 97.9 per cent specific for the prediction of maternal mortality. This ensures that no ‘at-risk’ woman misses attention, though many of them would survive by initiating appropriate management.

Overall, it was seen that there was a progressive worsening of outcomes with increasing scores. Participants with an aggregate score of ≥4.5 required CCU admission, while those with a score of ≥5.5 ended up as near-miss cases. Most maternal deaths occurred at an aggregate score ≥8.5. This highlights its potential to aid graded response that allows timely intervention and appropriate resource allocation.

RR emerged as the most significant parameter in our study. This is in coherence with Ryan et al8 and Singh et al7, who found RR to be the most important parameter, next only to temperature and conscious level, respectively. The findings of this study reinforce the importance of evaluating RR in all pregnant women, irrespective of their risk stratification.

A pregnant woman needs to provide for the developing foetus, and already has low respiratory reserves and is in a state of respiratory alkalosis. Hence, oxygen therapy should be initiated if her saturation falls to ≤96 per cent. FiO₂ required to maintain a saturation above 96 per cent, when triggered (1.2%), thus becomes an important indicator of deterioration.

Conscious level is a parameter that was rarely triggered (0.5%), but when triggered, it was of high value (relative risk: 17.7). Other studies have also reported it to be a significant parameter7,12. This is because a non-alert state reflects a low Glasgow Coma Scale (≤14) which signifies brain hypoxia and may result in permanent sequelae if timely interventions are not initiated.

While DBP appeared non-significant in our study (relative risk –1.4) and Paternina-Caicedo et al12 study which evaluated Carle’s OEWS, it was a significant parameter in studies by Singh et al6 (odds ratio: 0.64; relative risk: 6.6) and Singh et al7 (relative risk: 4.06) which employed CEMACH MEOWS. Paternina-Caicedo et al12 in their study had already pointed out this limitation of Carle’s OEWS. They found that using Carle’s method, a woman with only DBP <90 mmHg did not need further care or follow up. However, the non-parametric test used for this hypothesis showed that, while ignoring the cut-offs suggested by Carle et al11, DBP had a strong association with mortality12.

SBP and HR were comparable in most studies. Temperature remained the least significant parameter in most studies. However, in a study by Ryan et al8, β-coefficient for maximum temperature was the highest.

The strength of the present study lies in the fact that it was prospective in nature and conducted on pregnant women labour ward settings with a robust sample size. It evaluated Carle’s OEWS which is considered the scientific upgradation of the popular CEMACH MEOWS. OCCU admission was taken as the outcome measure, as ‘morbidity’ would overburden the tertiary centres and ‘mortality’ would not allow timely intervention.

However, Carle’s escalation pathway was not followed as this was an observational study in a tertiary care centre and so case management was based on the hospital protocol.

Overall, the results of our study suggest that Carle’s OEWS is a good predictor of maternal deterioration. Conscious level (0.5%) and FiO₂ required to maintain SpO₂ ≥96 per cent (1.2%) were rarely triggered, but when triggered, they triggered to the highest score of three, demanding immediate attention. DBP has poor predictability in Carle’s OEWS. Trigger values for DBP need to be reconsidered for any significant association as recommended in the present study and the previous study by Paternina-Caicedo et al12.

Further interventional studies are recommended in different population settings to determine the utility of Carle’s OEWS as a screening tool for predicting poor maternal outcomes using the escalation protocol suggested by Carle.