Temporal deep learning framework for retinopathy prediction in patients with type 1 diabetes

The adoption of electronic health records in hospitals has ensured the availability of large datasets that can be used to predict medical complications. The trajectories of patients in real-world settings are highly variable, making longitudinal data modeling challenging. In recent years, significant progress has been made in the study of deep learning models applied to time series; however, the application of these models to irregular medical time series (IMTS) remains limited. To address this issue, we developed a generic deep-learning-based framework for modeling IMTS that facilitates the comparative studies of sequential neural networks (transformers and long short-term memory) and irregular time representation techniques. A validation study to predict retinopathy complications was conducted on 1207 patients with type 1 diabetes in a French database using their historical glycosylated hemoglobin measurements, without any data aggregation or imputation. The transformer-based model combined with the soft one-hot representation of time gaps achieved the highest score: an area under the receiver operating characteristic curve of 88.65%, specificity of 85.56%, sensitivity of 83.33% and an improvement of 11.7% over the same architecture without time information. This is the first attempt to predict retinopathy complications in patients with type 1 diabetes using deep learning and longitudinal data collected from patient visits. This study highlighted the significance of modeling time gaps between medical records to improve prediction performance and the utility of a generic framework for conducting extensive comparative studies.

Prevalence of sexually transmitted infections in patients with schizophrenia in Morocco

OBJECTIVES

People with schizophrenia are likely to be overrepresented in high-risk categories for sexually transmitted infections (STIs). We aimed to determine the seroprevalence of HIV, HBV, HBC, and syphilis in these patients.

PATIENTS AND METHODS

Sociodemographic and clinical characteristics of 444 patients were collected, as well as risk factors for STIs. Serological tests were performed.

RESULTS

Lifetime seroprevalence rates were 0.0% for HIV, 1.6% for HBV, 0.9% for HBC, and 3.6% for syphilis. These figures are lower than those reported in the general population except for syphilis. Only 0.45% of patients were seropositive for both HBV and syphilis.

CONCLUSION

No higher rates of HBV, HBC, and HIV seroprevalence were observed compared with the general population, except for syphilis.

Deep Learning versus Conventional Machine Learning for Detection of Healthcare-Associated Infections in French Clinical Narratives

OBJECTIVE

The objective of this article was to compare the performances of health care-associated infection (HAI) detection between deep learning and conventional machine learning (ML) methods in French medical reports.

METHODS

The corpus consisted in different types of medical reports (discharge summaries, surgery reports, consultation reports, etc.). A total of 1,531 medical text documents were extracted and deidentified in three French university hospitals. Each of them was labeled as presence (1) or absence (0) of HAI. We started by normalizing the records using a list of preprocessing techniques. We calculated an overall performance metric, the F1 Score, to compare a deep learning method (convolutional neural network [CNN]) with the most popular conventional ML models (Bernoulli and multi-naïve Bayes, k-nearest neighbors, logistic regression, random forests, extra-trees, gradient boosting, support vector machines). We applied the hyperparameter Bayesian optimization for each model based on its HAI identification performances. We included the set of text representation as an additional hyperparameter for each model, using four different text representations (bag of words, term frequency-inverse document frequency, word2vec, and Glove).

RESULTS

CNN outperforms all other conventional ML algorithms for HAI classification. The best F1 Score of 97.7% ± 3.6% and best area under the curve score of 99.8% ± 0.41% were achieved when CNN was directly applied to the processed clinical notes without a pretrained word2vec embedding. Through receiver operating characteristic curve analysis, we could achieve a good balance between false notifications (with a specificity equal to 0.937) and system detection capability (with a sensitivity equal to 0.962) using the Youden's index reference.

CONCLUSIONS

The main drawback of CNNs is their opacity. To address this issue, we investigated CNN inner layers' activation values to visualize the most meaningful phrases in a document. This method could be used to build a phrase-based medical assistant algorithm to help the infection control practitioner to select relevant medical records. Our study demonstrated that deep learning approach outperforms other classification learning algorithms for automatically identifying HAIs in medical reports.

[Thrombosis of the inferior vena cava revealing primary antiphospholipid syndrome: a case report]

Thrombosis of the inferior vena cava is a rare clinical expression of primary antiphospholipid syndrome. This case clearly illustrates the clinical manifestations, work-up findings and management principles of thrombosis of the inferior vena cava in primary antiphospholipid syndrome before the stage of complication. The patient was a 24-year-old female with a history of recurrent deep venous thrombosis. She was admitted to the department of internal medicine for dyspnea and pain of the right hypochondria. Physical examination disclosed an edematous ascitic syndrome. The scanner coupled with Doppler ultrasonography showed thrombosis of the inferior vena cava. On these radiological findings, an immunological work-up was requested, which showed the presence of antiphospholipid antibodies.

Localized scleroderma: a series of 52 patients

BACKGROUND

Localized scleroderma also called morphea is a skin disorder of undetermined cause. The widely recognized Mayo Clinic Classification identifies 5 main morphea types: plaque, generalized, bullous, linear and deep. Whether each of these distinct types has a particular clinical course or is associated with some patient-related features is still unclear.

METHODS

We report here a retrospective series of patients with localized scleroderma with an attempt to identify features related to the type of lesion involved. The medical records of all patients with a diagnosis of localized scleroderma were reviewed by skilled practitioners. Lesions were classified according to the Mayo Clinic Classification. The relationship between each lesion type and various clinical features was tested by non-parametrical methods.

RESULTS

The sample of 52 patients included 43 females and 9 males. Median age at onset was 30 y (range 1-76). Frequencies of patients according to morphea types were: plaque morphea 41 (78.8%) (including morphea en plaque 30 (57.7%) and atrophoderma of Pasini-Pierini 11 (21.1%)), linear scleroderma 14 (26.9%). Nine patients (17.3%) had both types of localized scleroderma. Median age at onset was lower in patients with linear scleroderma (8 y (range 3-44)) than in others (36 y (range 1-77)) (p=0.0003). Head involvement was more common in patients with linear scleroderma (37.5%) than in other subtypes (11.1%) (p=0.05). Atrophoderma of Pasini-Pierini was never located at the head. Systemic symptoms, antinuclear antibodies and the rheumatic factor were not associated with localized scleroderma types or subtypes.

CONCLUSION

These results suggest that morphea types, in adults are not associated with distinct patient features except for age at disease onset (lower) and the localization on the head (more frequent), in patients with lesions of the linear type.

Neutrophil proteinase cathepsin G is proteolytically active on the human platelet glycoprotein Ib-IX receptor: characterization of the cleavage sites within the glycoprotein Ib alpha subunit

The proteolytic activity of the neutrophil serine-proteinase cathepsin G (CG) on platelet adherence receptors, the glycoprotein (GP) Ib-IX complex and the integrin alpha IIb beta 3, has been investigated. In the range 50 to 200 nmol/l, CG is a potent platelet agonist which induces shape change, granule exocytosis and aggregation. Investigation of the proteolysis of the receptors' subunits during the course of platelet activation by CG was performed by immunoblot analysis of platelet proteins using a panel of specific antibodies. Exposure of platelets for 3 min at 37 degrees C to CG at a concentration that induces full cell activation resulted in an extensive cleavage of the N-terminal region of the extracellular domain of GPIb alpha, the largest (relative molecular mass, M(r), 143,000) of the three subunits constituting the GPIb-IX complex. In contrast, no detectable proteolytic modification of the two other subunits, GPIb beta and GPIX, was detected. Similarly, we observed that neither of the two subunits of the alpha IIb beta 3 receptor were proteolytically modified by CG. Cleavage of GPIb alpha by CG leaves a remnant of the polypeptide chain with M(r) approx. 106,000 in the plasma membrane, while releasing into the extracellular milieu the N-terminal domain with M(r) in the range 40,000 to 46,000. N-terminal sequencing of the CG-derived fragments of GPIb alpha indicated that the Leu275-Tyr276 peptide bond was the primary cleavage site for this proteinase. Proteolysis of GPIb alpha was already detectable at concentrations of CG as low as 25 nmol/l, while with 200 nmol/l the cleavage was detected as soon as 10 s after exposure of platelets to the proteinase. Comparison of the kinetics and concentration dependency for the proteolysis of GPIb alpha and for the activation of platelets by CG showed that cleavage of the GPIb-IX receptor is an early event that accompanies exocytosis and aggregation. Quantitative evaluation of the conversion of GPIb alpha into its membrane fragment indicated that, under optimal conditions, a maximum of approx. 50% of the total GPIb alpha can be affected by proteolysis. However, this proteolysis was > 90% complete when platelets were in the presence of the potent antagonist prostacyclin, suggesting that cellular redistribution of the GPIb-IX receptor may also occur during activation by CG. These results thus indicate that the very early phase of platelet activation by CG is accompanied by extensive modifications in the structure and expression of the GPIb-IX receptor, an effect that might be of functional significance for the interaction of platelets with the vessel wall.

Regulation of the structure and activity of platelet adhesion receptors by leukocyte proteinases

Two major membrane receptors implicated in the adhesive properties of blood platelets are the GPIb-IX complex, a receptor for subendothelial von Willebrand factor, and the alpha IIb beta 3 integrin, the receptor for plasma fibrinogen. We have evaluated how the biological activities of these receptors can be potentially modulated through limited proteolysis when platelets are exposed to the serine-proteinases secreted by activated polymorphonuclear neutrophils, i.e., leukocyte elastase (EL) and cathepsin G (CG). CG can activate the alpha IIb beta 3 integrin through intracellular metabolic pathways, but has no direct proteolytic activity on the receptor subunits. By contrast, EL does not activate the platelet metabolism, but specifically cleaves a short peptide sequence within the alpha IIb subunit, and this cleavage occurs in parallel with an up-regulation of the activity of the fibrinogen receptor. On another hand, both EL and CG cleave the amino-terminal portion of the GPIb alpha subunit of the GPIb-IX receptor, eliminating the binding site for von Willebrand factor and diminishing the capacity of platelets to interact with this adhesion protein. Thus, neutrophil proteinases have the potential to regulate the activity of platelet adhesion receptors, and such experimental observations may prove to be relevant in vivo in various pathological conditions.