Evaluation of deep learning-based depression detection using medical claims data

Human accuracy in diagnosing psychiatric disorders is still low. Even though digitizing health care leads to more and more data, the successful adoption of AI-based digital decision support (DDSS) is rare. One reason is that AI algorithms are often not evaluated based on large, real-world data. This research shows the potential of using deep learning on the medical claims data of 812,853 people between 2018 and 2022, with 26,973,943 ICD-10-coded diseases, to predict depression (F32 and F33 ICD-10 codes). The dataset used represents almost the entire adult population of Estonia. Based on these data, to show the critical importance of the underlying temporal properties of the data for the detection of depression, we evaluate the performance of non-sequential models (LR, FNN), sequential models (LSTM, CNN-LSTM) and the sequential model with a decay factor (GRU-Δt, GRU-decay). Furthermore, since explainability is necessary for the medical domain, we combine a self-attention model with the GRU decay and evaluate its performance. We named this combination Att-GRU-decay. After extensive empirical experimentation, our model (Att-GRU-decay), with an AUC score of 0.990, an AUPRC score of 0.974, a specificity of 0.999 and a sensitivity of 0.944, proved to be the most accurate. The results of our novel Att-GRU-decay model outperform the current state of the art, demonstrating the potential usefulness of deep learning algorithms for DDSS development. We further expand this by describing a possible application scenario of the proposed algorithm for depression screening in a general practitioner (GP) setting-not only to decrease healthcare costs, but also to improve the quality of care and ultimately decrease people's suffering.

The epidemiology of patellar luxation in an insured Swedish dog population

Patellar luxation (PL) is a common orthopaedic condition in dogs. This study aimed to evaluate the incidence and cause-specific mortality rate, age at diagnosis, and risk factors for medial PL (MPL), lateral PL (LPL), and bidirectional PL (BPL). Other diagnoses in dogs with PL were also explored. The study population included just over 600,000 dogs insured by Agria Pet Insurance in Sweden (2011-2016). There were 2726 dogs with PL. Medial patellar luxation affected 90 % of the dogs with PL, followed by LPL (5.9 %), BPL (2.4 %), and unspecified PL (1.6 %). The median age at first diagnosis during the study period was 2.8 years for MPL, 2.7 years for LPL, and 1.5 years for BPL. In total, 168 (6.2 %) of the dogs with PL had cruciate ligament rupture. There were substantial breed-specific differences in the risk of PL: almost all breeds at increased risk of MPL were small-sized, while several of the breeds at increased risk of LPL were large-sized. The breeds at high risk of BPL varied in size. Females had an increased risk of MPL (RR 1.2, 95 % CI: 1.1-1.3, p < 0.001) and a decreased risk of LPL (RR 0.72, 95 % CI: 0.51-1.0, p = 0.042) compared to males. In total, 116 dogs were euthanised due to PL and the breeds with the highest risk of PL-related euthanasia were the Pyrenean mountain dog, Dogue de Bordeaux, and German pinscher. The median age for PL-related euthanasia was 2.2 years.

Rates of Complications After Appendectomy in Children and Adolescents: Pediatric Surgical Compared to General Surgical Hospitals

BACKGROUND

Appendectomies in children and adolescents are performed in Germany in pediatric surgical (PS) or general surgical hospitals (GS). The aim of this study is to evaluate whether the surgery in a PS or GS hospital has an influence on the postoperative course after appendectomy in children and adolescents.

MATERIALS AND METHODS

Nationwide routine data from children and adolescents aged 1-17 y insured by the Local Health Insurance Fund who underwent appendectomy between 2014 and 2016 were analyzed (cohort study). Descriptive statistics were calculated both overall and in the two groups (PS and GS). Patients were additionally examined by age (1-5, 6-12, and 13-17 y), treatment (laparoscopic, open surgical, and conversion), and appendicitis type (nonacute: K36/K37/K38/R10, acute simple: K35.30/K35.8, and acute complex: K35.2/K35.31/K35.32). The influence of surgeon specialization on 90-d secondary surgery and 90-d general complications was assessed by multiple logistic regression.

RESULTS

Altogether, 25,065 patients who underwent surgery in 83 PS and 906 GS hospitals were included. Logistic regression analysis revealed that PS was associated with a reduced risk of interventions in the 1-5- and 6-12-y age groups (odds ratio: 0.44, 0.62). Acute complex appendicitis, comorbidities, and open surgery significantly increased the risk for reintervention. PS was associated with an increased risk for complications in the 13-17-y age group (odds ratio: 1.66).

CONCLUSIONS

PS and GS hospitals provided safe appendectomies in children and adolescents with low reintervention and complication rates. PS hospitals demonstrated advantages for patients in the 1-5- and 6-12-y age groups and GS hospitals for patients 13-17 y.

Hip fracture risk in patients with burn injury: a retrospective cohort study in Taiwan

This work aimed to evaluate the hip fracture risk for patients with burn injury. A total of 16,430 patients with burn injury had an adjusted hazard ratio of 1.54 to encounter a hip fracture, compared with controls without the injury. These results encourage future studies focusing on mechanisms leading to fracture associated with burn injury.

INTRODUCTION

The relationship between burn injury and hip fracture risk is unclear. We conducted a retrospective cohort study to investigate this relationship.

METHODS

From insurance data of Taiwan, we identified a cohort with 16,430 burn patients in 2000-2010 and a comparison cohort of 65,716 persons without the history of burn, frequency matched by sex, age, and diagnosis date. Both cohorts were followed up to the end of 2011 to evaluate the risk of hip fracture.

RESULTS

Patients with burn injury were 1.62-fold more likely than comparisons to encounter a hip fracture (6.95 vs. 4.28 per 1000 person-years), with an adjusted hazard ratio (aHR) of 1.54 (95% confidence interval (CI) = 1.40-1.68). The fracture incidence increased with age and is slightly greater for women than for men in both cohorts. The fracture risk was greater for patients with burn in the eyes, face, and head with an incidence of 7.14 per 1000 person-years, or an aHR of 2.09 (95% CI = 1.53, 2.86). Diabetes and osteoporosis were also associated with an increased hip fracture risk.

CONCLUSION

Burn injury is associated with an increased risk of hip fracture. Diabetes and osteoporosis are associated with an enhanced risk.

A current life table and causes of death for insured dogs in Japan

The life expectancies and causes of death were evaluated in 299,555 dogs insured in Japan between 1 April 2010 and 31 March 2011, of which 4169 dogs died during this period. The overall life expectancy of dogs was 13.7 years. The probability of death was high in the first year of life, lowest in the second and third years, and increased exponentially after 3 years of age. The life expectancy was 13.8 years in the <5 kg body weight group, 14.2 years in the 5-10 kg body weight group, 13.6 years in the 10-20 kg body weight group, 12.5 years in the 20-40 kg body weight group and 10.6 years in the ≥40 kg body weight group. As body weight increases, life expectancy tended to decrease except in the <5 kg body weight group. The probability of death increased as dogs got older for most potential causes of death. Neoplasia resulted in the highest probability of death, especially in the large and giant breed groups. Cardiovascular system disorders were the second major cause of death and the toy group had a probability of death significantly higher than the other breed groups at age 12+.

Gynecological malignancy risk in colorectal cancer survivors: A population-based cohort study

PURPOSE

This study was carried out to assess the risk of gynecological malignancy in colorectal cancer survivors using a population-based retrospective cohort study.

METHOD

Using the National Health Insurance Research Database (NHIRD) of Taiwan, we identified 37,176 patients with colorectal cancer diagnosed in 1998-2009, aged 20 years and above, without other cancer history. We also randomly selected 148,700 women without any cancer in the comparison cohort, frequency matched by age and diagnosis date. Incidences and hazards of breast, cervix, endometrial and ovarian cancers were evaluated by 201l.

RESULTS

The overall incidence of the 4 types of gynecological cancer was 39.0% higher in colorectal cancer patients than in comparisons (2.99 vs. 2.14 per 1000 person-years) with an adjusted hazard ratio (HR) of 1.46 (95% confidence interval (CI) = 1.31-1.62). Breast cancer accounted for most subsequent cancer. The multivariable Cox method measured HR was the highest for endometrial cancer (3.40, 95% CI = 2.59-4.47) for the colorectal cohort relative to comparisons, followed by ovarian cancer and breast cancer, except cervix cancer. The risk of gynecological malignancies was apparently elevated for colorectal cancer survivors <50 years of age.

CONCLUSIONS

Follow-up measures are suggested for women with colorectal cancer for early detection and prevention of the subsequent gynecological malignancy.