Worth the paper it’s written on? A cross-sectional study of Medical Certificate of Stillbirth accuracy in the UK

Abstract Background The Medical Certificate of Stillbirth (MCS) records data about a baby’s death after 24 weeks of gestation but before birth. Major errors that could alter interpretation of the MCS were widespread in two UK-based regional studies. Methods A multicentre evaluation was conducted, examining MCS issued 1 January 2018 to 31 December 2018 in 76 UK obstetric units. A systematic case-note review of stillbirths was conducted by Obstetric and Gynaecology trainees, generating individual ‘ideal MCSs’ and comparing these to the actual MCS issued. Anonymized central data analysis described rates and types of error, agreement and factors associated with major errors. Results There were 1120 MCSs suitable for assessment, with 126 additional submitted data sets unsuitable for accuracy analysis (total 1246 cases). Gestational age demonstrated ‘substantial’ agreement [K = 0.73 (95% CI 0.70–0.76)]. Primary cause of death (COD) showed ‘fair’ agreement [K = 0.26 (95% CI 0.24–0.29)]. Major errors [696/1120; 62.1% (95% CI 59.3–64.9%)] included certificates issued for fetal demise at <24 weeks’ gestation [23/696; 3.3% (95% CI 2.2–4.9%)] or neonatal death [2/696; 0.3% (95% CI 0.1–1.1%)] or incorrect primary COD [667/696; 95.8% (95% CI 94.1–97.1%)]. Of 540/1246 [43.3% (95% CI 40.6–46.1%)] ‘unexplained’ stillbirths, only 119/540 [22.0% (95% CI 18.8–25.7%)] remained unexplained; the majority were redesignated as either fetal growth restriction [FGR: 195/540; 36.1% (95% CI 32.2–40.3%)] or placental insufficiency [184/540; 34.1% (95% CI 30.2–38.2)]. Overall, FGR [306/1246; 24.6% (95% CI 22.3–27.0%)] was the leading primary COD after review, yet only 53/306 [17.3% (95% CI 13.5–22.1%)] FGR cases were originally attributed correctly. Conclusion This study demonstrates widespread major errors in MCS completion across the UK. MCS should only be completed following structured case-note review, with particular attention on the fetal growth trajectory.


Introduction
Stillbirth is defined in the UK as a child born after 24 weeks' gestation without signs of life, 1 occurring in 3.9/ 1000 English and Welsh births in 2018. 2 In the Medical Certificate of Stillbirth (MCS), the issuer records the probable cause of death (COD); this is issued to parents to legally register their child's death. The hospital retains a 'counterfoil' summarizing the information recorded on the corresponding MCS. Whereas stillbirth registration can be deferred for 21, 42 or 365 days in Scotland, England/Wales and Northern Ireland, respectively, in practice registration is required prior to the body being buried or cremated. Therefore, the MCS is traditionally issued within days. 3 MCS data are collated nationally to report on stillbirth incidence, timing and cause. 4,5 An 'unexplained' MCS COD is legally acceptable in the UK, 6 with no external Coronial accuracy checks. 7 Two regional UK-based studies examined MCSs issued in North-West England during 2009 and 2015 8,9 comprising 213/3688 (5.8%) and 243/3147 (7.7%) of reported stillbirths in England and Wales during the study periods. 2 It was found that 74.2% and 53.6% of 2009 and 2015 stillbirths, respectively, were registered as 'unexplained' but only 18.3% and 13.9% remained unexplained after review (cases reviewed for adjudicated COD comprised 93.0% and 91.4% of total cases submitted, respectively). Major errors (those likely to affect MCS interpretation by family, healthcare professionals or statisticians 10 ) were found in 49% of MCSs in 2015; minor errors (those that do not affect MCS interpretation 10 ) were found in a further 25%.
Accurately recording stillbirth COD is important for families and local and national learning. The MCS is the first and often only information source for bereaved parents and future caregivers to guide future pregnancy or maternal health management, 11 as only a minority undergo post-mortem investigation. 12 Studies in adult death certification suggest that certifier profession and training may affect accuracy. 13,14 Data from MCSs also inform healthcare service strategy, funding, research and public health initiatives. 5 It is imperative to identify preventable stillbirths to aid future strategies to reduce deaths. This will support international governments in meeting the Every Newborn Action Plan 2030 targets. 15 We hypothesized that (i) the previous finding from regional studies, that a majority of MCSs have inaccuracies, would be replicated nationwide; (ii) major MCS error would be related to issuer profession and seniority.

Methods
The study is reported in accordance with STROBE guidelines (Supplementary Table S1, available as Supplementary data at IJE online). A total population sampling method was adopted.
The United Kingdom Audit and Research Collaborative in Obstetrics and Gynaecology (UKARCOG), 16 a traineeled organization, maintains regional representatives in all seven National Health Service Health Education England regions 17 and one representative each in Northern Ireland, Scotland and Wales. Representatives recruit local data collectors (doctors pursuing a career in obstetrics and gynaecology) to specific project working groups from every hospital in their region. For this project, no restrictions were placed on local data collectors' seniority/prior training. Local data collectors registered the individual project with their local governance processes.
Cases were identified via birth registers, MCS counterfoils and local perinatal mortality registers. In previous regional studies >90% of cases were identified via local perinatal mortality registers and MCS data were obtained from counterfoils in >75% of cases. 8,9 Inclusion criteria were birth during the study period of 1 January 2018 to 31 December 2018 and (i) MCS issuance or (ii) where an MCS should have been issued but was not. 18 In accordance with UK law, stillbirth following medical termination of pregnancy beyond 24 completed weeks' gestation was included. Potential cases were excluded if local data collectors were unable to verify eligibility for legal registration of birth (whether stillborn or live-born) against maternal medical records or in late miscarriage/neonatal death cases where no evidence of an MCS actually being issued was found. The 1-year study period was chosen to mitigate temporal changes in COD and workforce experience that may bias the study findings and for a realistic workload (estimated 2000-2500 cases based on national stillbirth rates and assuming data submitted from 80% of UK maternity units). 2,19 Data collection commenced in August 2019 after junior doctor hospital rotations, minimizing potential for local data collectors to self-audit.

Data collection
An electronic local data collection tool (Supplementary File S2, available as Supplementary data at IJE online) was adapted from previous studies, 8,9 piloted using case notes from one hospital (L.H.) and tested with fictional data combinations (M.R. and W.P.S.). Contemporaneous parental autopsy wishes were incompletely documented in maternal medical records, preventing accurate assessment against MCSs, and MCS autopsy status accuracy was therefore excluded from review. Local data collectors received written instructions for structured case review 20 and data collection tool use.
Data regarding the woman's background, pregnancy health prior to fetal death, care provided and information documented on the actual MCS were reviewed locally, limited to data available at MCS completion (contemporaneous data; Figure 1). 3,11,21 Delayed investigation results, defined as those not reported prior to the MCS being issued (including autopsy, placental histopathology, cytogenetics, thrombophilia screening and some microbiological tests), were disregarded unless reported prior to MCS completion, as these results could not have been known by the MCS issuer. An 'ideal MCS' was constructed by the data collector. Documented and ideal CODs were categorized according to the Relevant Condition at Death (ReCoDe) classification system (Table 1). 22 Conditions present but Figure 1 Typical timeline for reporting post-mortem investigation results in the UK. Reporting timeline based on full uptake of all recommended investigations, 11 including autopsy, and average reporting times; 3 days since delivery are shown along the bottom with territory-specific legal registration deadlines in brackets [Northern Ireland deadline for registration (365 days) not pictured in displayed timeline]. In practice, legal registration of stillbirth must be complete before burial or cremation; therefore, it is commonly completed within a week of delivery (indicated by bold arrow). The National Health Service aims to report 60% of autopsy investigations within 42 days of examination and 90% within 56 days, longer where specialist opinions are required. 22 Accounting for processing time and delays, many autopsy examinations can take up to 12 weeks for reporting; placental histopathology without autopsy is often reported by 6 weeks post-delivery. Similar timescales are anticipated in other international healthcare settings. In practice the true time from death to delivery is unknown and may, particularly in cases of co-twin demise, take several weeks. However, diagnosisto-delivery intervals of >1 week are unusual in singleton pregnancies. * indicates investigations that are not always indicated. Autopsy consent is optional ( $ ) and not required under Coronial jurisdiction in the UK.   not contributing to death were deemed 'associated' rather than 'causal'. Where multiple conditions were considered causal, their hierarchy (direct or indirect) was determined according to likely contribution to the infant's demise. Local data collectors assessed the accuracy ('accurate', 'inaccurate' or 'unknown') of MCS-reported infant sex, delivery date, birthweight, timing in relation to labour onset (termed 'accuracy data').
As in previous regional studies the following definitions were applied: lethal fetal abnormality (abnormality typically associated with 50% perinatal mortality), late miscarriage (fetal death in utero or birth without sign of life between 18 þ0 and 23 þ6 weeks' gestation) and placental insufficiency (placental weight <10th centile for gestation and fetal sex, 23 abnormal umbilical artery Doppler waveform, oligohydramnios without fetal abnormality or ruptured membranes, or documented small, infarcted, gritty or otherwise abnormal placenta at birth). 9 Oligohydramnios, abnormal umbilical artery Doppler, abnormal other Doppler (including but not limited to uterine artery, middle cerebral artery, Ductus venosus) and other conditions were defined according to local policy. 'Previous audit region' was hospitals included in prior North-West England regional audits of MCS accuracy.
Customized estimated fetal weight and individualized birthweight centiles were calculated using the Bulk Centile Calculator V.8.0.4 (Perinatal Institute, Birmingham, UK). In order to reduce potential for fetal growth restriction (FGR) over-diagnosis, gestation calculated from the date of fetal death in utero confirmation was used to calculate the birthweight centile, with an in-built 2-day gestation 'correction' within the centile calculator where fetal demise is indicated. FGR was defined according to the 2016 liveborn FGR consensus definition with an additional criterion: loss of >50 centiles in symphysio-fundal height (with no growth scans), provided birthweight <50th centile. 24 Abdominal circumference centiles were calculated according to Chitty et al. 25 The completing clinician seniority was categorized as junior (Registered Midwife Band 5/6, Obstetrics and Gynaecology speciality trainee Years 1-5 or equivalent) or senior (Registered Midwife Band 7þ, Obstetrics and Gynaecology speciality trainee Years 6þ, Consultant Obstetrician or equivalent).

Data review
Several quality-control measures were taken. Data collection form validation rules alerted local data collectors contemporaneously when inconsistent/impossible data were entered (e.g. date, gestation, age, weight or height outside plausible range). Summary data, after personal identifying data were removed, were analysed centrally for eligibility by two independent reviewers (L.H. and M.P.R.) who did not participate in data collection. Cases were excluded if eligibility could not be assured, e.g. simultaneously reporting no access to MCS data (therefore lacking MCS issuance evidence) and either unknown gestation with birthweight <500 g or possible signs of life at birth for the same case. 1 Missing values were not imputed. Where cases were excluded from accuracy assessment (all accuracy assessment data missing), included and excluded case characteristics were compared (Mann-Whitney U test and Chi 2 test). Where some (but not all) accuracy data were reported, cases were retained and missing data were assumed to be accurate. Local data collector assessments could not be validated against post-mortem findings (low uptake) or national perinatal mortality review findings (case/hospital anonymity), but have previously been shown to be similar. 9,26 Data analysis Primary analyses were major and minor error rates (defined as in Table 2), error-free rates, and primary COD and gestational age accuracy [Cohen's kappa coefficient (K)]. 27 Potentially identifiable data (including indirectly identifying data combinations such as date and timing in relation to delivery) were not submitted centrally. MCS counterfoils do not include infant sex. Thus preventing other accuracy field chance-adjusted agreements being calculated.
Major MCS error odds were calculated for issuer profession and seniority, using multilevel regression adjusting for clustering within hospitals. Based on previous regional studies, we estimated 99% power to detect a major error association of magnitude odds ratio (OR) <0.67 when comparing doctors to midwives, with 50% precision and 1:3 'exposure' to 'non-exposure'; power for other exposure/outcome pairs will vary. Exploratory secondary analyses compared individual condition/category incidences between actual and ideal MCS and major error association with maternal, healthcare and case characteristics but lack power to detect subtle relationships with lower exposure frequencies. Sensitivity analysis (complete records only) was conducted.

Study cohort
The UKARCOG National Evaluation of Accuracy of Stillbirth Certificates (NESTT) working group comprised 115 individuals from 76/158 (49.4%) UK healthcare trusts providing maternity services in 2018 (including data submission from at least one hospital from all 10 UKARCOG representative regions). Of 1270 cases submitted, 24 (1.9%) were excluded and 126 (9.9%) were unsuitable for accuracy analysis due to non-submission of any MCS accuracy data ( Figure 2) Cases assessed for MCS accuracy differed from those excluded by country (P < 0.0001) and level of maternity care provided in the hospital issuing the MCS (P ¼ 0.032), initial level of maternity care provided (P ¼ 0.011), antenatal assessment of fetal growth (P ¼ 0.014), birthweight centile (P ¼ 0.027) and diagnosisto-delivery interval (P ¼ 0.006) (Supplementary Table S3, available as Supplementary data at IJE online).
After adjustment for agreement by chance, agreement between MCS documented gestational age and the gestational age at fetal death in utero confirmation was 'substantial' (K ¼ 0.73; 95% CI 0.70-0.76). Corresponding agreement for primary COD was 'fair' (K ¼ 0.26; 95% CI 0.24-0.29). Factors associated with major MCS error are shown in Table 4; the MCS issuer profession and seniority were unrelated to major error odds (P > 0.05).  Errors in birthweight were not considered to be major, as if this resulted in failure to document the contribution of abnormal fetal growth, this would be covered under either incorrect primary cause of death (major error) or inclusion/exclusion of a contributory condition (not) present (minor error). Comparisons of accuracy were absolute; the magnitude of difference between true and documented values did not affect classification as a major or minor error, as these variables would not affect the interpretation of the certificate, except where indicated b in which case if the requirements for major error were met, this would be classified as a major error. . Several continuous variables associated with major error odds but fell short of pre-specified OR thresholds for potential significance at examined increments (BMI, parity and birthweight centile).

Sensitivity analysis
Full accuracy assessment was submitted in 181/1120 (16.2%) cases. Comparison between the sensitivity and remaining cohorts (data not shown) demonstrated a higher baseline antenatal care level (P < 0.0001) and more frequent escalation in care during pregnancy (P < 0.0001) vs     ] was associated with lower odds of major error for the first time. The previously reported associations between odds of major error and previous audit region, any previous stillbirth or late miscarriage, lethal congenital abnormality, pregnancy termination, birthweight centile 3-9.9 or placental weight centile <10 were no longer demonstrated in the sensitivity cohort. The continuous variables BMI and parity, but not number of previous stillbirths or birthweight centile, all remained associated with major error (full results in Supplementary Table S6, available as Supplementary data at IJE online).

Principal findings
MCS completion inaccuracies occurred irrespective of geographical location, with just one in five MCSs error-free. MCS issuer professional training was unrelated to major error odds. Hospitals in the previous audit region were less likely to issue a major error-containing MCS, possibly due to increased error awareness following regional initiatives. 8,9,20 Structured assessment enables non-expert reviewers to assign likely COD in most stillbirths. FGR remains frequently undocumented on MCSs, despite being the principal COD in a quarter of cases. Placental COD (including abruption and insufficiency) were also found to be frequently underreported as cause of death after case review.

Interpretation
Major and minor error and error-free rates, error types and chance-adjusted primary COD concordance in this geographically diverse cohort replicate prior regional UK MCS studies 8,9 and neonatal death certificate studies. 31 Our study reconfirms that stillbirth data accuracy is substantially worse than adult and child death certificate accuracy. [32][33][34][35] The USA and other countries report similar issues, 36-40 providing broader relevance for this UK study.
Case ascertainment was lower than in the MBRRACE-UK report. 12 Given the differential characteristics of cases excluded from accuracy assessment, it is possible that excluded cases may have been more likely to have remained unexplained compared with those included in the accuracy assessment. Hospitals not participating in this study may have more errors and be less motivated to improve MCS accuracy, thus biasing our analysis in favour of inflated Summarizing the adjusted odds ratios for major error (that which is likely to alter the interpretation of the Medical Certificate of Stillbirth by family, healthcare professionals or healthcare statisticians) by hypothesis testing variables (profession and seniority) and all exploratory variables found to be associated with odds of major error within the accuracy cohort (reference group as defined in table). All odds are adjusted for clustering within hospitals through multilevel regression. For each variable/category the number of records with major error and total number of records assessed in each category are displayed along with the odds ratio, 95% CI. Due to missing data and small numbers of records in certain categories, caution should be applied when interpreting odds ratios from less frequently reported variables/categories. Further, there was insufficient power to adjust for potential confounding between different exploratory variables. To see the full results of all examined variables in the accuracy cohort and within the sensitivity cohort, please see Supplementary  24 with the additional inclusion criteria of loss of >50 centiles in symphysio-fundal height in the absence of serial growth scans (provided that birthweight was below at least 50th centile). c Gestation and sex-adjusted placental weight centiles were calculated according to Thompson et al. 23 accuracy reporting. Similarly, the assumed accuracy of missing data (which was widespread) in this study is highly likely to be incorrect based on our sensitivity analyses. Together, our results likely overestimate true MCS accuracy nationwide. Nonetheless, MCS primary COD agreement (0.26) is unacceptable and lower than reported (between expert reviewers) in the Cause of Death and Associated Conditions perinatal classification system development (0.59), the only other perinatal mortalityspecific classification system with reported inter-observer reliability. 41 Although there is controversy regarding the 'best' perinatal death classification system, ReCoDe was chosen because it includes FGR and placental causes (leading UK preventable stillbirth causes and essential criteria of global stillbirth classification systems) as direct CODs. 42 COD assessment guidelines for stillbirth may improve accuracy. Certificates issued from the previous audit region, where regional guidelines for structured case review and MCS completion are provided, 20 demonstrated fewer major errors. Specific birthweight centile and growth trajectory assessment increased FGR identification, which may not be obvious from birthweight. Many stillborn FGR infants in this study had exceeded recommended gestationrelated delivery thresholds 5 and had excellent neonatal survival chances. 43 Some such FGR deaths might have been preventable; this should be continually audited.
MCS issuer professional training appears unrelated to the major error likelihood, mirroring the pattern in adult death certification. 44 It is notable that some factors protective against major MCS error are those associated with increased risk of stillbirth such as previous stillbirth or congenital abnormalities; this may reflect increased awareness and contemporaneous documentation of factors predisposing to stillbirth in such women. 11 The persistent association of birthweight <3rd centile and FGR diagnosis (confirmed post-birth) with increased major error odds likely reflects specific FGR diagnosis assignment instruction provision to local data collectors, whereas original MCS issuers had no such instruction. Similarly, local definitions for other conditions may have introduced variation in assignment across the cohort. However, assignment frequency in many conditions changed after case review.

Strengths
We have demonstrated that using a structured case review, non-expert reviewers are able to assign a COD for the majority of stillbirths without specific training. Utilizing a similar method and the same perinatal death classification system as used in previous North-West England studies allows comparisons over time; this study provides greater accuracy assessment depth compared with previous studies by examining secondary COD and related condition data. Kappa statistical analysis allows discernment between agreement by chance and systematic agreement.
Over-attribution to FGR has been minimized using a restricted definition, birthweight centile calculations that adjust for post-mortem in utero retention and considering all conditions' relevant contributions, rather than adhering to the ReCoDe hierarchical structure. This allowed alternative primary COD assignment (e.g. placental abruption) where appropriate, even in FGR infants. 45 Use of the stillborn FGR definition would have been ideal, but this was published after data collection commenced 46 and relies on autopsy results, which were unavailable contemporaneously as explained above. Restricting case-note review to contemporaneous data ensures the adjudicated COD could have been contemporaneously assigned. Unless stillbirth certification/registration systems are redesigned to allow amendment for delayed post-mortem investigation, as in Wisconsin (USA), 47 individual and nationally reported stillbirth CODs will remain dependent on the individual certifier and case-review quality.

Limitations
Low sample size with high missing data rates impaired detection of associations between patient/healthcare variables and major error odds, and prevented multivariable adjustment for potential confounding. Future studies should be appropriately powered to confirm or refute the associations we found. Caution should particularly be applied when interpreting sensitivity analysis data (83% cases excluded), which was underpowered to detect/confirm potential clinically important relationships. Despite this, the results remain meaningful and informative. Although included case characteristics (birthweight, gestation, maternal ethnicity) are similar to those in national reports, 12,28-30 it is not possible to compare the reported and unreported case characteristics as patient identity protection outside the direct clinical care team was paramount. Finally, distinction between major and minor errors may be semantic. However, those errors defined as 'minor' would not independently result in management change or alter retrospective evaluation of antemortem management and therefore deserve less attention.

Conclusion
We identified nationwide stillbirth certification and COD assignment deficiencies, confirming our first hypothesis. This reinforces the need for independent, national perinatal surveillance processes such as MBRRACE-UK. However, the lack of evidence that MCS accuracy relates to issuer profession and seniority, combined with the lower odds of major error in previous audit region hospitals, suggests that basic tools, guidance and awareness raising among individual MCS certifiers and hospitals may greatly help to improve data quality. Correct stillbirth cause classification is crucial for families and society; when 'unexplained', conditions' true perinatal mortality contributions are uncounted and preventative strategies cannot be appropriately targeted. Without this, the UK and other countries will fail to meet international stillbirth reduction targets. Therefore, it is imperative to ensure MCS accuracy and to undertake a continuous feedback and performance improvement cycle. Until stillbirth certification systems allow COD amendments, individual and nationally reported stillbirth CODs will remain dependent on the individual certifier and case-review quality.