Predicting low cognitive ability at age 5 using machine learning methods and birth cohort data

Background

Early intervention is essential to address disparities in cognitive development. Current developmental screening will not detect the vast majority of children who go on to have below average cognitive ability at school age. In this study, we applied the random forest (RF) algorithm, a highly interpretable machine learning method, to birth-cohort data to train a model to predict low cognitive ability at 5 years of age using perinatal features.

Methods

Data was from 1,070 participants in the Irish population-based BASELINE birth cohort. A RF model was trained to predict an intelligence quotient (IQ) score <90 at age 5 years using a broad selection of maternal, infant, birth, and sociodemographic features, all of which could be easily measured at a population level in the perinatal period. Feature importance was examined using mean decrease in Gini impurity, mean decrease in accuracy, and mean minimal depth. Recursive feature elimination was used to develop a parsimonious model. Internal validation was performed using 10-fold cross validation repeated 5 times.

Results

The most predictive features for low cognitive ability at 5 years of age were the total years of maternal schooling, infant Apgar score at 1 minute, socioeconomic index, maternal BMI, and units of alcohol consumed in the first trimester. A parsimonious RF model based on 11 features showed excellent predictive ability, with a sensitivity of 0.89 and a specificity of 0.98, providing a foundation suitable for external validation in an unseen cohort.

Conclusions

Machine learning approaches to large existing datasets can provide accurate feature selection to improve risk prediction. Further validation of this model is required in additional cohorts, representative of the general population. Accurate risk prediction can facilitate targeted screening and intervention.

Key messages

• The application of machine learning to large population-based data can improve feature selection and accuracy in risk prediction models.

• Accurate risk prediction may enable early intervention to address disparities in cognitive development. Individual interventions must occur in conjunction with population level policy changes.

P-432 Biologic use during conception and pregnancy and its impact on fetal and maternal outcomes: systematic review and meta-analysis

Study question

To determine pregnancy outcomes in women with chronic inflammatory disease exposed to biologics during conception and pregnancy.

Summary answer

Meta-analysis of 11172 pregnancies exposed to biologic medications show no evidence of harm for the fetus or the mother.

What is known already

Chronic inflammatory diseases (CIDs) are a group of autoimmune diseases which affect between 5-7% of the population and include rheumatoid arthritis (RA), psoriatic arthritis and inflammatory bowel disease (IBD). Many CIDs have a female preponderance and are often associated with activity during reproductive years. Biologic medications, specifically the TNF-α inhibitors, have become increasingly prevalent in the treatment of chronic inflammatory disease (CID) in women of childbearing age.

Study design, size, duration

PubMed and EMBASE databases January 1998-July 2021.Peer reviewed, English language cohort, case-control, cross-sectional studies, and case series which contained original data. P- diagnosis of CID pregnancy. I- Biologic medication. C- diagnosis of CID without treatment with biologics and a CID free population. O- fetal:congenital malformations, preterm delivery (<37 weeks), severe neonatal infection requiring hospitalisation, low birth weight (<2.5Kg) and small for gestational age (<10th Centile). Maternal: severe maternal infection requiring hospitalisation, miscarriage and pre-eclampsia.

Participants/materials, setting, methods

Two authors independently conducted data extraction. A meta-analysis of proportions using a random-effects model was used to pool outcomes. Linear regression analysis was used to compare the mean of proportions of outcomes across exposure groups using the ‘treated’ group as the reference category. All studies were evaluated using an appropriate quality assessment tool described by McDonald et al. The GRADE approach was used to assess the overall certainty of evidence.

Main results and the role of chance

35 studies, 11172 pregnancies, were eligible for inclusion. Analysis showed pooled proportions for congenital malformations: treated 0.04(95% CI 0.03-0.04; I2 77) vs disease matched 0.04(0.03-0.05. I2 86) p = 0.238. Preterm delivery treated 0.04(0.10-0.14. I2 88) vs disease matched 0.10(0.09-0.12. I2 87) p = 0.250. Severe neonatal infection: treated 0.05(0.03-0.07. I2 88) vs disease matched 0.05(0.02-0.07. I2 94) p = 0.970. Low birth weight: treated 0.10(0.07-0.12. I2 93) vs disease matched 0.08(0.07-0.09. I2 0) p = 0.241. The pooled Miscarriage: treated 0.13(0.10-0.15. I2 77) vs disease matched 0.08(0.04-0.11. I2 5) p = 0.078. Pre-eclampsia; treated 0.01(0.01-0.02. I2 0) vs disease matched 0.01(0.00-0.01. I2 0). p = 0.193. No statistical differences in proportions were observed.

Limitations, reasons for caution

GRADE certainty of findings were low to very low.

Wider implications of the findings

We demonstrated comparable pregnancy outcomes in pregnancies exposed to biologics, disease matched controls and CID free pregnancies using the GRADE approach.

Trial registration number

NA

The Practice of Vitamin D Deficiency Assessment in a Geriatric Day Hospital

There has been no clear consensus on the assessment and treatment of vitamin D deficiency prior to the publication of the National Osteoporosis Society (NOS) Vitamin D Guideline in 2014. The aim of our study was to assess the practice in a medicine for the older person day hospital setting relative to this guideline. A 6-month retrospective analysis of all new patients who attended service from January to July 2013 was carried out. Seventy-six patients were included in the final analysis. Mean age was 83 years. 39 (51%) patients had sufficient levels while 37 (49%) patients had insufficient levels; 14 (19%) being inadequate and 23 (30%) deficient. Eighteen patients who had insufficient levels were subsequently prescribed supplements; 13 (72%) received vitamin D3 in combination with calcium while 5 (28%) received vitamin D3 alone. Based on the findings of this study, we have made some recommendations and adopted the guideline.

Focal amplification of the androgen receptor gene in hormone-naive human prostate cancer

BACKGROUND

Androgen receptor (AR)-gene amplification, found in 20-30% of castration-resistant prostate cancer (CRPCa) is proposed to develop as a consequence of hormone-deprivation therapy and be a prime cause of treatment failure. Here we investigate AR-gene amplification in cancers before hormone deprivation therapy.

METHODS

A tissue microarray (TMA) series of 596 hormone-naive prostate cancers (HNPCas) was screened for chromosome X and AR-gene locus-specific copy number alterations using four-colour fluorescence in situ hybridisation.

RESULTS

Both high level gain in chromosome X (≥4 fold; n=4, 0.7%) and locus-specific amplification of the AR-gene (n=6, 1%) were detected at low frequencies in HNPCa TMAs. Fluorescence in situ hybridisation mapping whole sections taken from the original HNPCa specimen blocks demonstrated that AR-gene amplifications exist in small foci of cells (≤ 600 nm, ≤1% of tumour volume). Patients with AR gene-locus-specific copy number gains had poorer prostate cancer-specific survival.

CONCLUSION

Small clonal foci of cancer containing high level gain of the androgen receptor (AR)-gene develop before hormone deprivation therapy. Their small size makes detection by TMA inefficient and suggests a higher prevalence than that reported herein. It is hypothesised that a large proportion of AR-amplified CRPCa could pre-date hormone deprivation therapy and that these patients would potentially benefit from early total androgen ablation.

High-density tissue microarrays from prostate needle biopsies

Background

Formalin-fixed prostate biopsies are frequently the only tissue collected at the time of prostate cancer diagnosis. There is therefore a requirement for techniques that allow the use of these prostate biopsy specimens in a high-throughput analysis of immunohistochemical and fluorescence-in-situ-hybridisation-detected biomarkers.

Methods

The authors have previously described methods that allow tissue microarray (TMA) construction from prostate biopsies. Here, we describe significant technical innovations that provide an easier and more robust system of biopsy–TMA construction.

Results and discussion

The TMAs produced are of a high density (up to 104 cores each, 8×13) and allow a multiplex analysis of biomarkers in the context of clinical trials.

Resource utilisation for syncope presenting to an acute hospital Emergency Department

BACKGROUND

Syncope is a common clinical problem accounting for up to 6% of hospital admissions. Little is known about resource utilisation for patients admitted for syncope management in Ireland.

AIM

To determine the utilisation of resources for patients admitted for syncope management.

METHODS

Single centre observational case series of consecutive adult patients presenting to an acute hospital Emergency Department with syncope over a 5-month period.

RESULTS

Two-hundred and fourteen of 18,898 patients (1.1%) had a syncopal episode, 110 (51.4%) of whom were admitted. Mean length of stay was 6.9 days. Sixty-four of these admissions were deemed unnecessary by retrospective review when compared to ESC guidelines. Eighty-five (77.3%) admitted patients had cardiac investigations and 56 (51%) had brain imaging performed.

CONCLUSIONS

Syncope places a large demand on overstretched hospital resources. Most cases can be managed safely as an outpatient and to facilitate this, hospitals should develop outpatient Syncope Management Units.

An improved method for constructing tissue microarrays from prostate needle biopsy specimens

Background:

Prostate cancer diagnosis is routinely made by the histopathological examination of formalin fixed needle biopsy specimens. Frequently this is the only cancer tissue available from the patient for the analysis of diagnostic and prognostic biomarkers. There is, therefore, an urgent need for methods that allow the high-throughput analysis of these biopsy samples using immunohistochemical (IHC) markers and fluorescence in situ hybridisation (FISH) analysis based markers.

Methods:

A method that allows the construction of tissue microarrays (TMAs) from diagnostic prostate needle biopsy cores has previously been reported. However, the technique only allows the production of low-density biopsy TMAs with a maximum of 20 cores per TMA. Here two methods are presented that allow the rapid and uniform production of biopsy TMAs containing between 54 and 72 biopsy cores. IHC and FISH techniques were used to detect biomarker status.

Results:

Biopsy TMAs were constructed from prostate needle biopsy specimens taken from 102 patients entered into an active surveillance trial and 201 patients in a radiotherapy trial. The detection rate for cancer in slices of these biopsy TMAs was 66% and 79% respectively. Slices of a biopsy TMA prepared from biopsies from active surveillance patients were used to detect multiple IHC markers and to score TMPRSS2-ERG fusion status in a FISH-based assay.

Conclusions:

The construction of biopsy TMAs provides an effective method for the multiplex analysis of IHC and FISH markers and for their assessment as prognostic biomarkers in the context of clinical trials.

Bovine germinal vesicle oocyte and cumulus cell proteomics

Germinal vesicle (GV) breakdown is fundamental for maturation of fully grown, developmentally competent, mammalian oocytes. Bidirectional communication between oocytes and surrounding cumulus cells (CC) is essential for maturation of a competent oocyte. However, neither the factors involved in this communication nor the mechanisms of their actions are well defined. Here, we define the proteomes of GV oocytes and their surrounding CC, including membrane proteins, using proteomics in a bovine model. We found that 4395 proteins were expressed in the CC and 1092 proteins were expressed in oocytes. Further, 858 proteins were common to both the CC and the oocytes. This first comprehensive proteome analysis of bovine oocytes and CC not only provides a foundation for signaling and cell physiology at the GV stage of oocyte development, but are also valuable for comparative studies of other stages of oocyte development at the molecular level. Furthermore, some of these proteins may represent molecular biomarkers for developmental potential of oocytes.

Anticonvulsant use in elderly patients in long-term care units

BACKGROUND

Elderly patients in long-term care units are frailer than their community-dwelling peers and may be more at risk from toxic side-effects of anticonvulsant medication at standard doses.

AIM

To examine the prescribing of anticonvulsants to patients in elderly care units.

METHODS

Drug prescription sheets and case notes were reviewed. Serum anticonvulsant concentration, renal and liver profiles and albumin level were measured.

RESULTS

Anticonvulsants were prescribed to twice as many male as female patients (32 vs 14%; p<0.03) and to 33% of those younger than 80 years of age versus 10% of those aged 80 years or older (p<0.0002). No patient had significant hypoalbuminaemia and routine measurement of serum anticonvulsant concentration did not indicate an alteration of dosage.

CONCLUSIONS

Anticonvulsants appear to be well tolerated in these patients. The younger age of those receiving anticonvulsants is inadequately explained by the characteristics of the patient cohort and may reflect a shift towards a younger age in patients requiring anticonvulsants due to increased mortality in this group.

Primary cell cultures isolated from Penaeus monodon prawns

We have devised a cell culture system for Penaeus monodon prawn cells that uses a defined synthetic medium. Organs were removed from adult prawns ranging in size from 13--19cm rostrum-to-telson length. Cultures consisted of either a blend of hematopoietic and lymphoid cells or ovarian cells. The cells divide rapidly in culture, doubling on average once per week for 5 to 6 weeks. These cultures continue to survive for at least 5 months but the rates of cell division are low after the first 5--6 weeks. In the literature, unicellular eukaryotic marine organisms such as chytrids may contaminate marine cell cultures. In some cases these eukaryotic contaminants may be difficult to distinguish from prawn cells unless detailed ultrastructure or characteristic developmental stages such as zoospores can be observed. Alternatively, we prepared molecular probes from repeated DNA sequences 100--400 bp in length in the P. monodon genome. These species-specific probes were hybridised to genomic DNA from cell culture to confirm proliferation of P. monodon cells in our cultures.

Characterization of 65 bovine microsatellites

Microsatellites or simple sequence repeat (SSR) polymorphisms are used widely in the construction of linkage maps in many species. High levels of polymorphism coupled with the ease of analysis of the polymerase chain reaction (PCR) have resulted in this type of maker being one of the most widely used for genetic analysis. In this paper we describe 58 polymorphic bovine microsatellites that were isolated from insert size selected bovine genomic libraries. Primer sequences, number of alleles, and heterozygosity levels in cattle reference families are reported. Chromosomal locations for 47 of these microsatellites as well as for 7 previously described systems derived from entries in the Genbank or EMBL databases have been determined. The markers map to 24 syntenic or chromosomal locations. Polymorphic bovine microsatellites were estimated to occur, on average, every 320 kb, and there is no evidence of clustering in the genome. Thirty of the bovine-derived microsatellite systems gave specific and polymorphic products in sheep, adding to the number of useful markers in that species.

Septicemia in long-term jugular hemodialysis catheters; eradicating infection by changing the catheter over a guidewire

By convention, septicemia occurring from an infected vascular catheter is treated with antibiotics and removal of the catheter. This approach, used with surgically implanted long-term catheters would be expected to result in loss of the vascular access site. During a 57 month period, we treated 21 episodes of septicemia secondary to infection of long term indwelling double lumen jugular venous catheters in our hemodialysis unit. Seventeen of 21 episodes were managed successfully by changing the catheter over a guidewire, thus preserving the access site. No relapse was observed after the antibiotic therapy was stopped. Only four patients required complete removal of the catheter and subsequent use of another site.