Evaluation of safety and performance of CentaFlow™ in the assessment of fetal growth restriction – a randomized trial and prospective cohort study

Key message

1. Centaflow was safe to use on pregnant women.

2. There was no evidence that Centaflow can be used to detect fetal growth restriction.

Introduction

Fetal growth restriction (FGR) affects approximately 3 % of all pregnancies [1], 2] and is a condition related to abnormal placental development when no chromosomal aberrations or immunization are responsible [3], 4]. FGR is associated with an increased risk of intrauterine death, preterm birth, and a series of long-term adverse consequences including cardiovascular and metabolic disease in both mother and child and altered neurodevelopmental outcomes in the child [5], [6], [7], [8], [9]. However, FGR is not easily identified prenatally, as in vivo assessment of the placental development and function have not yet been convincing. Therefore, clinical assessment and measurement of the symphysis fundal height as well as the fetal heartbeat using a stethoscope, in locations with no universal 3rd-trimester ultrasound screening, have been the reference standard [10]. Nevertheless, these widely used screening methods have low sensitivities [1], which increases the risk of intrauterine deaths and neonatal complications that could potentially be reduced with correct antenatal identification of FGR [11], 12].

The existence of maternal vascular murmurs (MVM) from the uteroplacental unit and the association with FGR was previously reported by Riknagel et al. based on small cross-sectional studies [13], [14], [15]. They tested a device called Centaflow™ (CF), which is a wireless patch with microphones placed on the lower abdomen of the pregnant women to assess the presence of MVM [13], [14], [15]. Their results showed that the majority of cases of FGR do exhibit MVM, but on the contrary, not all healthy pregnancies or pregnancies with fetuses <10th percentile with normal umbilical artery flow, present MVM [14]. Due to the size and type of previous studies, the possible clinical implications still need to be investigated. Moreover, the overall safety of use of CF on the mother and the child needs to be determined on a larger scale.

Hence, the overall aim was to evaluate the sensitivity and specificity of CF, and secondary to evaluate the safety of the CF device. We hypothesized that the device is safe to use and has a higher sensitivity than standard care.

Materials and methods

This was a sponsor-initiated multicenter randomized controlled clinical trial (RCT) with a termination of the randomization after enough women had been included to evaluate the safety of CF. After the termination of the randomization, the study proceeded as a prospective multicenter study. Performance was evaluated in both the randomized part and the prospective part of the study. The study was conducted between 03/06/2020 and 21/03/2022 at Copenhagen University Hospital Rigshospitalet, Copenhagen and Regional Hospital Central Jutland, Viborg, and Regional Hospital West Jutland, Herning. Women +18 years of age with a singleton pregnancy more than 12 weeks’ gestation up until delivery were invited to participate. Exclusion criteria were fetal abnormality known to affect fetal growth or affect the birth weight (including malformations, genetic disorders, and immunization) diagnosed at any given time prenatally and up to 12 days postpartum. In the prospective study, only high-risk pregnancies, defined as an estimated fetal weight <15 %, were invited to participate to include women with growth restricted babies to evaluate the sensitivity of CF. This part also included women from Copenhagen University Hospital Herlev, Copenhagen and Copenhagen University Hospital North Zealand, Hillerod.

In Denmark, we have tax-paid universal healthcare, hence the prenatal screening program and obstetric care are free of charge for the individual. All pregnant women are offered two prenatal ultrasound screening exams – one in the 1st trimester screening for trisomies and one in the 2nd trimester screening for malformations with an attendance of more than 95 % of the pregnant population [16]. Moreover, women are offered 5–7 antenatal visits with a midwife and three visits with a general practitioner. If any complications are suspected, the patients are referred to the obstetric department for further evaluation. This is nationwide standard. All women were recruited through antenatal visits at the midwife or at the ultrasound clinic.

The Centaflow device

The system is not yet CE-marked under the medical device regulation (MDR) and is investigated as a Class IIa medical device. The indication of use is the detection of murmurs (MVM) as a sign of intra-arterial turbulence in the maternal side of the placenta. The CF System consists of an active device with two sets of bilateral microphones (one set placed on the inner side and another on the outer side) and seven electrocardiogram (ECG) electrodes mounted to an adhesive patch (Figure 1, image). The CF device is placed on the lower abdomen of the pregnant woman. A computer tablet is connected to control the recording where the recorded audio and ECG is stored and sent to a central server. The ECG electrodes measure the maternal pulse to segment the sound recording from the R-wave in the ECG and are not used for other purposes. The body-contact microphones were developed with the Cardio-technology group at Aalborg University (Department of Health Science & Technology) and the Acoustics Group (Department of Electronic Systems) and have previously been tested [15].

Figure 1:

Image of the Centaflow device.

Training of investigators and trial personnel included investigational device usage, electronic case report form completion, medical device Good Clinical Practice (GCP) (ISO 14155), and trial personnel responsibilities.

The CF recorded for 3 min. A data complier function provided information of a) sufficient data or b) insufficient data. If b) was displayed, a new option for recording was provided, and by acceptance, an additional three-minute recording proceeded. After this, a new evaluation was performed by the compiler function and provided information of either a) or b). If the data were still insufficient, the examination was registered as incomplete.

All recordings were manually assessed by a minimum of two assessors. The recordings went through several steps of interpretation and analyses: 1) assessment of the quality; 2) evaluation of the recording by two independent, and blinded assessors (blinded for estimated fetal size and outcome); 3) and consensus meeting when disagreement occurred between the assessors. Step 2 was categorized as 1–3) murmur; 4) no murmur; and 5) uncertain. A murmur was classified as persistent systolic high frequency like described in previous studies [13], [14], [15]. A positive signal was defined as a signal classified between 1 and 3, where the signals were grouped 1, 2, or 3 depending on noise.

Results

A total of 1,601 pregnant women were included: 1,418 women were randomized to either SC+CF (n=703) or SC (n=715), and 183 women were included in the prospective study after the termination of the randomization (Figure 2, Flowchart). Hence, 886 were evaluated by CF. Baseline characteristics for those randomized are shown in Table 1.

Figure 2:

Flowchart of study population.

Of the 886 participants evaluated by CF, 501 were screened positive at least once (56.5 %). A total of 123 children were FGR (<3rd percentile) at birth, of which 88 were evaluated by CF, and 117 had a clinical assessment of the estimated fetal weight. We needed to include 133 pregnancies with true FGR according to our sample size calculation, however, due to an error in the equation in one of the sponsor’s systems used when including patients, we ended up with a population of 123 FGR cases.

In a total of 18 women, no clinical assessment of the fetal weight was registered. Table 2 shows the calculated sensitivity and specificity of CF. There was no evidence of any association between CF and FGR (p=0.192). The CF device gave a sensitivity of FGR of 50 % with a false positive rate of 57 %. Table 3 shows the sensitivity and specificity of clinical assessment of FGR. There was a sensitivity for FGR of 31.6 % with a false positive rate of 4.6 %.

Figure 3 shows the ROC-curves for the clinical assessment of the estimated fetal weight and symphysis-fundal height with areas under the curve of 0.82 and 0.73, respectively. For the same false-positive rate as CF, McNemar’s test showed that SC was superior to CF (p=0.014) (Supplementary Table 1).

Figure 3:

ROC-curves illustrating the performance of the clinical assessment by the midwife (A) and the symphysis-fundal height (B).

In the CF group, two patients experienced AEs as redness after removal of the adhesive patch which resolved spontaneously (0.3 %). The distribution of reported SAEs can be seen in Supplementary Table 2. The SAEs covered abnormal prenatal flow measurements, anhydramnion, polyhydramnion, elevated liver enzymes, hypertension, preeclampsia, HELLP syndrome, maternal infection, pulmonary embolism, shortened cervix, premature rupture of membranes/preterm premature rupture of membranes, intrapartum fever, placental abruptio, instrumental delivery, emergency cesarean section, peripartum hemorrhage, asphyxia, 3rd degree rupture of perineum, preterm birth (<37 weeks’ gestation), endometritis, mastitis, post-partum urinary retention, malformation, neonatal apnea, neonatal hypoglycemia, neonatal icterus, neonatal immunization, neonatal infection, respiratory distress syndrome, neonatal weight loss, foetus mors (n=5), and neonatal death (n=1). There were significantly more SAEs in the SC group compared to the CF group (p<0.001). There were two cases of intrauterine fetal death in the CF group and three cases in the SC group. The case of neonatal death was in the CF group. None of the SAEs were found to be associated with the use of CF.

Discussion

In this study, we evaluated the performance and the safety of CF for the first time in a large randomized controlled study (safety+performance) and a prospective study (performance). Overall, we found that CF is safe with no risk of serious adverse events, and only 0.3 % experienced minor skin problems (redness) from the adhesive patch. We found no evidence that CF can be used as a predictor of FGR, as the CF had a low sensitivity of 50 % and a specificity of 43 %. The positive predictive value was very low (8.8 %), though possibly affected by the low incidence of FGR, while the negative predictive value was acceptable (88.6 %). Thus, we were unable to replicate previous findings regarding the clinical performance of CF.

The performance of CF to detect FGR has previously been evaluated in one study [14]. Riknagel et al. performed a cross-sectional study over three months, including singleton pregnancies between 28- and 34-weeks’ gestation at a single center in Denmark. They investigated the correlation between MVM obtained from the CF device and estimated fetal weight<10th percentile along abnormal Doppler flow. They analyzed 63 pregnant women of which 25 were estimated to be fetal growth restricted based on ultrasound. The sensitivity for the outcome of interest was 36 %, specificity 79 %, and screen positive rate 27 %. The sensitivity and specificity improved markedly when the outcome was combined with abnormal umbilical artery PI (sensitivity 85 % and specificity 88 %). Our current results show a higher sensitivity, possibly due to the definition of FGR in the current study (<3rd-percentile), but a much lower specificity.

There are several possible reasons for the performance results including 1) The interpretation of the CF signals; 2) the design of the study to include patients in the entire third trimester; and 3) the definition of the outcome. This study was planned with manual annotations of the CF signals, i.e., a person listened to the recordings and categorized the sounds/murmurs. Despite efforts to minimize variability, this method implies a relative subjectivity in the evaluation of the sound signals, which introduces bias. It seems possible that it is more difficult for a person to reject abnormality than accept it, which could partly explain the low positive predictive value. Moreover, it is also possible that listening for a murmur is not the correct use of the CF. It is likely that more detailed, systematic, and objective signal analysis and assessment of the sound signals could improve the performance of CF. The study also included women through the entire third trimester, however, it has not been established at which gestational age the performance of CF is best. It may be independent of gestational age, though it appears more likely that the signal changes throughout pregnancy, similar to changes in the uterine artery Doppler flow observed at different gestational ages [19]. Lastly, the outcome was defined as FGR. The CF device can possibly detect adverse blood flow in the placenta, hence possibly detect poor placenta-associated outcomes not limited to FGR – and sometimes not related to FGR. Consequently, the appropriate outcome for CF could be adverse outcome associated with placental insufficiency.

Though the performance of CF was poor in detecting FGR, we found that the clinical assessment of the fetal weight performed by midwives performed well with an AUC of 0.82. The sensitivity was not remarkable, but the specificity of more than 95 % shows how accurately the Danish midwives can identify fetuses with normal growth. The results are better than what has been previously reported on the performance of clinical assessment to identify FGR [20].

The study has limitations. The sample size calculation estimated a need for 133 pregnancies with true FGR. However, due to a systematic error (error in the equation) in one of the systems used when including patients, we ended up with a population of 123 FGR cases. This is unfortunate, but we do not believe it would have changed any results. Also, due to the manual annotations of the signal, we believe that the results from this study should be interpreted with great caution. Moreover, we do not know the implications of the location of the placenta in the annotation of signals. This should be explored further in the future.

Though this study failed to show the possible potential of CF to detect FGR, we, as clinicians, still see the potential in this new biomarker. Despite the limitations, this study has demonstrated that the CF signal analysis and categorizations need to be reevaluated, and time windows of usage should be explored to determine the potential benefit of CF for FGR detection. Hence, future studies need to include 1) a new approach to assess the CF signals, i.e., through machine learning, and 2) more placenta-associated outcomes, i.e., preeclampsia and not only FGR. Moreover, the sounds of the normal pregnancy at low risk needs to be ascertained as well as the changes in the signal over time. By investigating and establishing these parameters, a more solid foundation could be laid to determine the possible usage of this new biomarker – alone or in combination with existing biomarkers.

In conclusion, we found that CF is safe to use, the performance of CF to detect FGR was poor. However, as there is still room for great improvement in both the characterization and analysis of the signal, we have not yet ruled out the possibility of CF as a new biomarker in the detection or prediction of placenta-associated adverse pregnancy outcomes.