Development and external validation of machine learning algorithms for postnatal gestational age estimation using clinical data and metabolomic markers.

BACKGROUND: Accurate estimates of gestational age (GA) at birth are important for preterm birth surveillance but can be challenging to obtain in low income countries. Our objective was to develop machine learning models to accurately estimate GA shortly after birth using clinical and metabolomic data. METHODS: We derived three GA estimation models using ELASTIC NET multivariable linear regression using metabolomic markers from heel-prick blood samples and clinical data from a retrospective cohort of newborns from Ontario, Canada. We conducted internal model validation in an independent cohort of Ontario newborns, and external validation in heel prick and cord blood sample data collected from newborns from prospective birth cohorts in Lusaka, Zambia and Matlab, Bangladesh. Model performance was measured by comparing model-derived estimates of GA to reference estimates from early pregnancy ultrasound. RESULTS: Samples were collected from 311 newborns from Zambia and 1176 from Bangladesh. The best-performing model accurately estimated GA within about 6 days of ultrasound estimates in both cohorts when applied to heel prick data (MAE 0.79 weeks (95% CI 0.69, 0.90) for Zambia; 0.81 weeks (0.75, 0.86) for Bangladesh), and within about 7 days when applied to cord blood data (1.02 weeks (0.90, 1.15) for Zambia; 0.95 weeks (0.90, 0.99) for Bangladesh). CONCLUSIONS: Algorithms developed in Canada provided accurate estimates of GA when applied to external cohorts from Zambia and Bangladesh. Model performance was superior in heel prick data as compared to cord blood data.

HIV-associated vaginal microbiome and inflammation predict spontaneous preterm birth in Zambia.

A Lactobacillus-deficient, anaerobe-rich vaginal microbiome has been associated with local inflammation and spontaneous preterm birth (sPTB), but few studies have assessed this association in the setting of HIV. We performed metagenomic sequencing and inflammatory marker assays on vaginal swabs collected in pregnancy. We grouped samples into 7 metagenomic clusters (mgClust) using the non-redundant VIRGO catalogue, and derived inflammatory scores by factor analysis. Of 221 participants, median Shannon diversity index (SDI) was highest in HIV+ with detectable viral load (1.31, IQR: 0.85-1.66; p < 0.001) and HIV+ with undetectable virus (1.17, IQR: 0.51-1.66; p = 0.01) compared to HIV- (0.74, IQR: 0.35-1.26). Inflammatory scores positively correlated with SDI (+ 0.66, 95%CI 0.28, 1.03; p = 0.001), highest among anaerobe-rich mgClust2-mgClust6. HIV was associated with predominance of anaerobe-rich mgClust5 (17% vs. 6%; p = 0.02) and mgClust6 (27% vs. 11%; p = 0.002). Relative abundance of a novel Gardnerella metagenomic subspecies > 50% predicted sPTB (RR 2.6; 95%CI: 1.1, 6.4) and was higher in HIV+ (23% vs. 10%; p = 0.001). A novel Gardnerella metagenomic subspecies more abundant in women with HIV predicted sPTB. The risk of sPTB among women with HIV may be mediated by the vaginal microbiome and inflammation, suggesting potential targets for prevention.

Maternal HIV Infection and Spontaneous Versus Provider-Initiated Preterm Birth in an Urban Zambian Cohort.

OBJECTIVE: We investigated the effect of maternal HIV and its treatment on spontaneous and provider-initiated preterm birth (PTB) in an urban African cohort. METHODS: The Zambian Preterm Birth Prevention Study enrolled pregnant women at their first antenatal visit in Lusaka. Participants underwent ultrasound, laboratory testing, and clinical phenotyping of delivery outcomes. Key exposures were maternal HIV serostatus and timing of antiretroviral therapy initiation. We defined the primary outcome, PTB, as delivery between 16 and 37 weeks' gestational age, and differentiated spontaneous from provider-initiated parturition. RESULTS: Of 1450 pregnant women enrolled, 350 (24%) had HIV. About 1216 (84%) were retained at delivery, 3 of whom delivered <16 weeks. Of 181 (15%) preterm deliveries, 120 (66%) were spontaneous, 56 (31%) were provider-initiated, and 5 (3%) were unclassified. In standardized analyses using inverse probability weighting, maternal HIV increased the risk of spontaneous PTB [RR 1.68; 95% confidence interval (CI): 1.12 to 2.52], but this effect was mitigated on overall PTB [risk ratio (RR) 1.31; 95% CI: 0.92 to 1.86] owing to a protective effect against provider-initiated PTB. HIV reduced the risk of preeclampsia (RR 0.32; 95% CI: 0.11 to 0.91), which strongly predicted provider-initiated PTB (RR 17.92; 95% CI: 8.13 to 39.53). The timing of antiretroviral therapy start did not affect the relationship between HIV and PTB. CONCLUSION: The risk of HIV on spontaneous PTB seems to be opposed by a protective effect of HIV on provider-initiated PTB. These findings support an inflammatory mechanism underlying HIV-related PTB and suggest that published estimates of PTB risk overall underestimate the risk of spontaneous PTB.

Maternal HIV, antiretroviral timing, and spontaneous preterm birth in an urban Zambian cohort: the role of local and systemic inflammation.

OBJECTIVE: To assess plasma and vaginal inflammation in three antenatal groups (HIV-uninfected women, HIV-infected women entering care on preconceptional ART, and HIV-infected women not on preconceptional ART) and whether these measures are associated with spontaneous preterm birth (sPTB). DESIGN: Case--control study nested within a pregnancy cohort in Lusaka, Zambia. METHODS: We analyzed 11 pro-inflammatory and two anti-inflammatory markers in 207 women with paired plasma and vaginal specimens collected between 16 and 20 gestational weeks. Among 51 HIV-infected women, we repeated the assays in 24-34-week samples. We used confirmatory factor analysis to create inflammation scores and compared them among the three groups. RESULTS: At baseline, HIV-infected women not on ART had higher vaginal pro-inflammatory scores than HIV-uninfected women [mean 0.37 (95% CI -0.06 to 0.80) vs. -0.02 (-0.32 to 0.27), P = 0.02]. In repeat testing, women not on preconceptional ART had an increase in vaginal inflammation between the baseline and 24-34-week visits compared with those continuing preconceptional ART [mean 0.62 (95% CI -0.80 to 4.20) vs. -0.07 (-2.78 to 2.11), P = 0.04]. In multivariate analyses, baseline vaginal inflammation predicted sPTB (aOR 1.5; 95% CI 1.0-2.3; P = 0.02). Plasma inflammation did not differ by HIV or ART exposure and was not associated with sPTB. CONCLUSION: Women not receiving ART at entry into pregnancy care had more vaginal inflammation than women entering on treatment. They also experienced an increase in vaginal inflammation between the two sampling timepoints, possibly as a consequence of ART initiation. Vaginal (but not systemic) inflammation was associated with sPTB and offers a potential mechanistic insight into this important adverse birth outcome.

Highly diverse anaerobe-predominant vaginal microbiota among HIV-infected pregnant women in Zambia.

Vaginal dysbiosis has been shown to increase the risk of some adverse birth outcomes. HIV infection may be associated with shifts in the vaginal microbiome. We characterized microbial communities in vaginal swabs collected between 16-20 gestational weeks in the Zambian Preterm Birth Prevention Study to investigate whether HIV and its treatment alter the microbiome in pregnancy. We quantified relative abundance and diversity of bacterial taxa by whole-genome shotgun sequencing and identified community state types (CST) by hierarchical clustering. Associations between exposures-HIV serostatus (HIV+ vs HIV-) and preconceptional ART (ART+ vs ART-)-and microbiome characteristics were tested with rank-sum, and by linear and logistic regression, accounting for sampling by inverse-probability weighting. Of 261 vaginal swabs, 256 (98%) had evaluable sequences; 98 (38%) were from HIV+ participants, 55 (56%) of whom had preconceptional ART exposure. Major CSTs were dominated by: L. crispatus (CST 1; 17%), L.] iners (CST 3; 32%), Gardnerella vaginalis (CST 4-I; 37%), G. vaginalis & Atopobium vaginae (CST 4-II; 5%), and other mixed anaerobes (CST 4-III; 9%). G. vaginalis was present in 95%; mean relative abundance was higher in HIV+ (0.46±0.29) compared to HIV- participants (0.35±0.33; rank-sum p = .01). Shannon diversity was higher in HIV+/ART+ (coeff 0.17; 95%CI (0.01,0.33), p = .04) and HIV+/ART- (coeff 0.37; 95%CI (0.19,0.55), p < .001) participants compared to HIV-. Anaerobe-dominant CSTs were more prevalent in HIV+/ART+ (63%, AOR 3.11; 95%CI: 1.48,6.55, p = .003) and HIV+/ART- (85%, AOR 7.59; 95%CI (2.80,20.6), p < .001) compared to HIV- (45%). Restricting the comparison to 111 women in either CST 3 (L. iners dominance) or CST 1 (L. crispatus dominance), CST 3 frequency was similar in HIV- (63%) and HIV+/ART- participants (67%, AOR 1.31; 95%CI: (0.25,6.90), p = .7), but higher in HIV+/ART+ (89%, AOR 6.44; 95%CI: (1.12,37.0), p = .04). Pregnant women in Zambia, particularly those with HIV, had diverse anaerobe-dominant vaginal microbiota.

HIV serostatus, viral load, and midtrimester cervical length in a Zambian prenatal cohort.

OBJECTIVE: To evaluate whether maternal HIV serostatus and plasma viral load (VL) are associated with midtrimester cervical length (CL). METHODS: The Zambian Preterm Birth Prevention Study (ZAPPS) is an ongoing prospective cohort that began enrolling in Lusaka in August 2015. Pregnant women undergo ultrasound to determine gestational age and return for CL measurement at 16-28 weeks. We evaluated crude and adjusted associations between dichotomous indicators and short cervix (≤2.5 cm) via logistic regression, and between VL and CL as a continuous variable via linear regression. RESULTS: This analysis includes 1171 women enrolled between August 2015 and September 2017. Of 294 (25.1%) HIV-positive women, 275 (93.5%) had viral load performed close to CL measurement; of these, 148 (53.8%) had undetectable virus. Median CL was 3.6 cm (IQR 3.5-4.0) and was similar in HIV-infected (3.7 cm, IQR 3.5-4.0) versus uninfected (3.6 cm, IQR 3.5-4.0) participants (P=0.273). The odds of short CL were similar by HIV serostatus (OR 0.64; P=0.298) and detectable VL among those infected (OR 2.37, P=0.323). We observed no association between log VL and CL via linear regression (-0.12 cm; P=0.732). CONCLUSION: We found no evidence of association between HIV infection and short CL.

Improving preterm newborn identification in low-resource settings with machine learning.

BACKGROUND: Globally, preterm birth is the leading cause of neonatal death with estimated prevalence and associated mortality highest in low- and middle-income countries (LMICs). Accurate identification of preterm infants is important at the individual level for appropriate clinical intervention as well as at the population level for informed policy decisions and resource allocation. As early prenatal ultrasound is commonly not available in these settings, gestational age (GA) is often estimated using newborn assessment at birth. This approach assumes last menstrual period to be unreliable and birthweight to be unable to distinguish preterm infants from those that are small for gestational age (SGA). We sought to leverage machine learning algorithms incorporating maternal factors associated with SGA to improve accuracy of preterm newborn identification in LMIC settings. METHODS AND FINDINGS: This study uses data from an ongoing obstetrical cohort in Lusaka, Zambia that uses early pregnancy ultrasound to estimate GA. Our intent was to identify the best set of parameters commonly available at delivery to correctly categorize births as either preterm (<37 weeks) or term, compared to GA assigned by early ultrasound as the gold standard. Trained midwives conducted a newborn assessment (<72 hours) and collected maternal and neonatal data at the time of delivery or shortly thereafter. New Ballard Score (NBS), last menstrual period (LMP), and birth weight were used individually to assign GA at delivery and categorize each birth as either preterm or term. Additionally, machine learning techniques incorporated combinations of these measures with several maternal and newborn characteristics associated with prematurity and SGA to develop GA at delivery and preterm birth prediction models. The distribution and accuracy of all models were compared to early ultrasound dating. Within our live-born cohort to date (n = 862), the median GA at delivery by early ultrasound was 39.4 weeks (IQR: 38.3-40.3). Among assessed newborns with complete data included in this analysis (n = 468), the median GA by ultrasound was 39.6 weeks (IQR: 38.4-40.3). Using machine learning, we identified a combination of six accessible parameters (LMP, birth weight, twin delivery, maternal height, hypertension in labor, and HIV serostatus) that can be used by machine learning to outperform current GA prediction methods. For preterm birth prediction, this combination of covariates correctly classified >94% of newborns and achieved an area under the curve (AUC) of 0.9796. CONCLUSIONS: We identified a parsimonious list of variables that can be used by machine learning approaches to improve accuracy of preterm newborn identification. Our best-performing model included LMP, birth weight, twin delivery, HIV serostatus, and maternal factors associated with SGA. These variables are all easily collected at delivery, reducing the skill and time required by the frontline health worker to assess GA. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT02738892.

Adverse birth outcomes and their clinical phenotypes in an urban Zambian cohort.

Background: Few cohort studies of pregnancy in sub-Saharan Africa use rigorous gestational age dating and clinical phenotyping. As a result, incidence and risk factors of adverse birth outcomes are inadequately characterized. Methods: The Zambian Preterm Birth Prevention Study (ZAPPS) is a prospective observational cohort established to investigate adverse birth outcomes at a referral hospital in urban Lusaka. This report describes ZAPPS phase I, enrolled August 2015 to September 2017. Women were followed through pregnancy and 42 days postpartum. At delivery, study staff assessed neonatal vital status, birthweight, and sex, and assigned a delivery phenotype. Primary outcomes were: (1) preterm birth (PTB; delivery <37 weeks), (2) small-for-gestational-age (SGA; <10 th percentile weight-for-age at birth), and (3) stillbirth (SB; delivery of an infant without signs of life). Results: ZAPPS phase I enrolled 1450 women with median age 27 years (IQR 23-32). Most participants (68%) were multiparous, of whom 41% reported a prior PTB and 14% reported a prior stillbirth. Twins were present in 3% of pregnancies, 3% of women had short cervix (<25mm), 24% of women were HIV seropositive, and 5% were syphilis seropositive. Of 1216 (84%) retained at delivery, 15% were preterm, 18% small-for-gestational-age, and 4% stillborn. PTB risk was higher with prior PTB (aRR 1.88; 95%CI 1.32-2.68), short cervix (aRR 2.62; 95%CI 1.68-4.09), twins (aRR 5.22; 95%CI 3.67-7.43), and antenatal hypertension (aRR 2.04; 95%CI 1.43-2.91). SGA risk was higher with twins (aRR 2.75; 95%CI 1.81-4.18) and antenatal hypertension (aRR 1.62; 95%CI 1.16-2.26). SB risk was higher with short cervix (aRR 6.42; 95%CI 2.56-16.1). Conclusio ns: This study confirms high rates of PTB, SGA, and SB among pregnant women in Lusaka, Zambia. Accurate gestational age dating and careful ascertainment of delivery data are critical to understanding the scope of adverse birth outcomes in low-resource settings.

Overview of Nonoperative Blunt Splenic Injury Management with Associated Splenic Artery Pseudoaneurysm.

The delayed development of splenic artery pseudoaneurysm (SAP) can complicate the nonoperative management of splenic injuries. We sought to determine the utility of repeat imaging in diagnosing SAP in patients managed nonoperatively without angioembolization. We hypothesized that a significant rate of SAPs would be found in this population on repeat imaging. Patients undergoing nonoperative splenic injury management from January 2011 to June 2015 were queried from the trauma registry. Rates of repeat imaging, angioembolization, readmission, and SAP development were analyzed. Further, subanalyses investigating the incidence of SAP in patients managed nonoperatively without angioembolization were conducted. A total of 133 patients met inclusion criteria. Repeat imaging rate was 40 per cent, angioembolization rate was 26 per cent, and readmission rate was 6 per cent. Within the study population, nine SAPs were found (8/9 in patients with splenic injury grade ≥III). Of these nine SAPs, three (33%) were identified on initial scans and embolized, whereas six (67%) were found on repeat imaging in patients not initially receiving angioembolization. Splenic injuries are typically managed nonoperatively without serious complications. Our results suggest patients with splenic injuries grade ≥III managed nonoperatively without angioembolization should have repeat imaging within 48 hours to rule out the possibility of SAP.

Vena Cava Filter Use in Trauma and Rates of Pulmonary Embolism, 2003-2015.

IMPORTANCE: Vena cava filter (VCF) placement for pulmonary embolism (PE) prophylaxis in trauma is controversial. Limited research exists detailing trends in VCF use and occurrence of PE over time. OBJECTIVE: To analyze state and nationwide temporal trends in VCF placement and PE occurrence from 2003 to 2015 using available data sets. DESIGN, SETTING, AND PARTICIPANTS: A retrospective trauma cohort study was conducted using data from the Pennsylvania Trauma Outcome Study (PTOS) (461 974 patients from 2003 to 2015), the National Trauma Data Bank (NTDB) (5 755 095 patients from 2003 to 2014), and the National (Nationwide) Inpatient Sample (NIS) (24 449 476 patients from 2003 to 2013) databases. MAIN OUTCOMES AND MEASURES: Temporal trends in VCF placement and PE rates, filter type (prophylactic or therapeutic), and established predictors of PE (obesity, pregnancy, cancer, deep vein thrombosis, major procedure, spinal cord paralysis, venous injury, lower extremity fracture, pelvic fracture, central line, intracranial hemorrhage, and blood transfusion). Prophylactic filters were defined as VCFs placed before or without an existing PE, while therapeutic filters were defined as VCFs placed after a PE. RESULTS: Of the 461 974 patients in PTOS, the mean (SD) age was 47.2 (26.4) and 61.6% (284 621) were men; of the 5 755 095 patients in NTDB, the mean age (SD) was 42.0 (24.3) and 63.7% (3 666 504) were men; and of the 24 449 476 patients in NIS, the mean (SD) age was 58.0 (25.2) and 49.7% (12 160 231) were men. Of patients receiving a filter (11 405 in the PTOS, 71 029 in the NTDB, and 189 957 in the NIS), most were prophylactic VCFs (93.6% in the PTOS, 93.5% in the NTDB, and 93.3% in the NIS). Unadjusted and adjusted temporal trends for the PTOS and NTDB showed initial increases in filter placement followed by significant declines (unadjusted reductions in VCF placement rates, 76.8% in the PTOS and 53.3% in the NTDB). The NIS demonstrated a similar unadjusted trend, with a slight increase and modest decline (22.2%) in VCF placement rates over time; however, adjusted trends showed a slight but significant increase in filter rates. Adjusted PE rates for the PTOS and NTDB showed significant initial increases followed by slight decreases, with limited variation during the declining filter use periods. The NIS showed an initial increase in PE rates followed by a period of stagnation. CONCLUSIONS AND RELEVANCE: Despite a precipitous decline of VCF use in trauma, PE rates remained unchanged during this period. Taking this association into consideration, VCFs may have limited utility in influencing rates of PE. More judicious identification of at-risk patients is warranted to determine individuals who would most benefit from a VCF.

A bitter pill to swallow: dysphagia in cervical spine injury.

BACKGROUND: Dysphagia is a common complication after cervical spine trauma with spinal cord injury. We sought to characterize the prevalence of dysphagia within a total cervical spinal injury (CSI) population, considering the implications of spinal cord injury status and age on dysphagia development. We hypothesized that while greater rates of dysphagia would be found in geriatric and spinal cord-injured subgroups, all patients presenting with CSI would be at heightened risk for swallowing dysfunction. METHODS: All trauma admissions to a level II trauma center from January 2010 to April 2014 with CSI were retrospectively reviewed. CSI was classified as any ligamentous or cervical spinous fracture with or without cord injury. Patients failing a formal swallow evaluation were considered dysphagic. The implications of dysphagia development on age and spinal cord injury status were assessed in univariate and multivariate analyses. RESULTS: A total of 481 patients met study inclusion criteria, of which 123 (26%) developed dysphagia. Within the dysphagic subpopulation, 90 patients (73%) were geriatric, and 23 (19%) sustained spinal cord injury. The dysphagic subpopulation was predominantly free from spinal cord injury (81%). Multivariate analyses found age (adjusted odds ratio: 1.06; 95% confidence interval 1.04-1.07; P < 0.001) and spinal cord injury (adjusted odds ratio: 2.69; 95% confidence interval 1.30-5.56; P = 0.008) to be significant predictors of dysphagia development. CONCLUSIONS: Despite spinal cord-injured patients being at increased risk for dysphagia, most of the dysphagic subpopulation was free from spinal cord injury. Geriatric and CSI patients with or without cord injury should be at heightened suspicion for dysphagia development.

Hyponatremia as a fall predictor in a geriatric trauma population.

INTRODUCTION: Approximately one in three older adults fall each year, resulting in a significant proportion of geriatric traumatic injuries. In a hospital with a focus on geriatric fall prevention, we sought to characterize this population to develop targeted interventions. As mild hyponatremia, defined as a serum sodium <135meq/L, has been reported to be associated with falls, unsteadiness and attention deficits, we hypothesized that hyponatremia is associated with falls in our geriatric trauma population. METHODS: Gender, age, pre-existing conditions (cardiac disease, diabetes, hematologic disorder, liver disease, malignancy, musculoskeletal disorder, neurological disorder, obesity, psychiatric disorder, pulmonary disease, renal disease, thyroid disease), mechanism of injury and admitting serum sodium level were queried for all geriatric trauma admissions from 2008 to 2011. Mechanism of injury was coded as falls admissions and non-falls admissions. Admitting serum sodium levels were coded as hyponatremic (<135mmol/L) and not hyponatremic (≥135mmol/L). RESULTS: Of the 2370 geriatric trauma admissions during the study period, there were 1841 (77.7%) falls admissions and 293 (12.4%) patients who were hyponatremic. Gender, age, neurological disorder, hematologic disorder, and hyponatremia were found to be significant predictors of falls in both univariate and multivariable analyses. CONCLUSION: Hyponatremic patients are significantly more likely to be admitted for a fall than non-hyponatremic patients, when adjusting for age, neurological disorder, and hematologic disorder. Consequently, hyponatremia identification and management should be an integral part of any geriatric trauma fall prevention programme. Additionally, if hyponatremia is found during a geriatric fall workup, it should be corrected prior to discharge and closely monitored by a primary care physician to prevent recurrent episodes of falls.