Cost-sensitive ordinal classification methods to predict SARS-CoV-2 pneumonia severity.

OBJECTIVE: To study the suitability of costsensitive ordinal artificial intelligence-machine learning (AIML) strategies in the prognosis of SARS-CoV-2 pneumonia severity. MATERIALS & METHODS: Observational, retrospective, longitudinal, cohort study in 4 hospitals in Spain. Information regarding demographic and clinical status was supplemented by socioeconomic data and air pollution exposures. We proposed AI-ML algorithms for ordinal classification via ordinal decomposition and for cost-sensitive learning via resampling techniques. For performance-based model selection, we defined a custom score including per-class sensitivities and asymmetric misprognosis costs. 260 distinct AI-ML models were evaluated via 10 repetitions of 5×5 nested cross-validation with hyperparameter tuning. Model selection was followed by the calibration of predicted probabilities. Final overall performance was compared against five well-established clinical severity scores and against a 'standard' (non-cost sensitive, non-ordinal) AI-ML baseline. In our best model, we also evaluated its explainability with respect to each of the input variables. RESULTS: The study enrolled n = 1548 patients: 712 experienced low, 238 medium, and 598 high clinical severity. d = 131 variables were collected, becoming d ' = 148 features after categorical encoding. Model selection resulted in our best-performing AI-ML pipeline having: a) no imputation of missing data, b) no feature selection (i.e. using the full set of d ' features), c) 'Ordered Partitions' ordinal decomposition, d) cost-based reimbalance, and e) a Histogram-based Gradient Boosting classifier. This best model (calibrated) obtained a median accuracy of 68.1% [67.3%, 68.8%] (95% confidence interval), a balanced accuracy of 57.0% [55.6%, 57.9%], and an overall area under the curve (AUC) 0.802 [0.795, 0.808]. In our dataset, it outperformed all five clinical severity scores and the 'standard' AI-ML baseline. DISCUSSION & CONCLUSION: We conducted an exhaustive exploration of AI-ML methods designed for both ordinal and cost-sensitive classification, motivated by a real-world application domain (clinical severity prognosis) in which these topics arise naturally. Our model with the best classification performance exploited successfully the ordering information of ground truth classes, coping with imbalance and asymmetric costs. However, these ordinal and cost-sensitive aspects are seldom explored in the literature.

Health services research in patients with breast cancer (CAMISS-prospective): study protocol for an observational prospective study.

BACKGROUND: Though breast cancer remains a major health problem, there is a lack of information on health care provided to patients with this disease and associated costs. In addition, there is a need to update and validate risk stratification tools in Spain. Our purpose is to evaluate the health services provided for breast cancer in Spain, from screening and diagnosis to treatment and prognosis. METHODS: Prospective cohort study involving 13 hospitals in Spain with a follow-up period of up to 5 years after diagnostic biopsy. Eligibility criteria: Patients diagnosed with breast cancer between April 2013 and May 2015 that have consented to participate in the study. DATA COLLECTION: Data will be collected on the following: pre-intervention medical history, biological, clinical, and sociodemographic characteristics, mode of cancer detection, hospital admission, treatment, and outcomes up to 5 years after initial treatment. Questionnaires about quality of life (EuroQoL EQ-5D-5 L, the European Organization For Research And Treatment Of Cancer Core Quality Of Life Questionnaire EORTC QLQ-C30 join to the specific breast cancer module (QLQ-BR23), as well as Hospital Anxiety and Depression Scale were completed by the patients before the beginning of the initial treatment and at the end of follow-up period, 2 years later. The end-points of the study were changes in health-related quality of life, recurrence, complications and readmissions at 2 and 5 years after initial treatment. STATISTICAL ANALYSIS: Descriptive statistics will be calculated and multivariate models will be used where appropriate to adjust for potential confounders. In order to create and validate a prediction model, split validation and bootstrapping will be performed. Cost analysis will be carried out from the perspective of a national health system. DISCUSSION: The results of this coordinated project are expected to generate scientifically valid and clinically and socially important information to inform the decision-making of managers and the authorities responsible for ensuring equality in care processes as well in health outcomes. For clinicians, clinical prediction rules will be developed which are expected to serve as the basis for the development of software applications. TRIAL REGISTRATION: NCT02439554 . Date of registration: May 8, 2015 (retrospectively registered) .

Economic Evaluation of Health Services Costs During Pandemic Influenza A (H1N1) Pdm09 Infection in Pregnant and Non-Pregnant Women in Spain.

BACKGROUND: The healthcare and socio-economic burden resulting from influenza A (H1N1) pdm09 in Spain was considerable. Our aim was to estimate and compare the management (resource utilization) and economic healthcare impact in an at-risk group of unvaccinated pregnant women with an unvaccinated group of non-pregnant woman of childbearing age (15-44 yr old). METHODS: We addressed this question with a longitudinal, observational, multicentre study. Inputs were the requirements in managing both groups of women. Outcome measures were healthcare costs. Direct healthcare (including medical utilisation, prescriptions of antivirals, medication, diagnostic tests, and hospitalisation) costs and indirect (productivity loss) costs were considered. Unit of cost was attributed to the frequency of health service resources utilisation. The mean cost per patient was calculated in this group of women. RESULTS: We found that the influenza clinical pattern was worse in non-pregnant women as they had a high medical risk of 20.4% versus 6.1% of pregnant women. Non-pregnant required more antipyretics and antibiotics, and needed more health service resource utilisation (338 medical visits in non-pregnant women vs. 42 in pregnant women). The total cost of non-pregnant women was higher (€4,689.4/non-pregnant and €2,945.07/pregnant). CONCLUSIONS: Cost per (H1N1) pdm09 was lower for pregnant women, probably due to more preventive measures adopted for their protection in Spain. The highest costs were incurred by hospitalisations/day and work absenteeism for non-pregnant than for pregnant women. These data will allow better future pandemic influenza planning.

Eating disorders: predictors of change in the quality of life of caregivers.

Eating disorders (EDs) have a strong impact on the quality of life (QoL) of caregivers. This study explores the longitudinal changes in the QoL of ED caregivers at 1 (T1; n=109) and 2 years (T2; n=32) follow-up, and identifies predictors of improvement at 1-year follow-up. ED outpatients also completed a battery of tests at T1 (n=69) and T2 (n=11). Multivariate hierarchical linear mixed models were used, in which the caregivers' 1-year changes in QoL, anxiety and depression scores were analysed as the dependent variables. Predictor variables were the caregivers' perception of burden, perception of the severity of the patient's ED, kinship data, and the ED person's scores on QoL, anxiety depression and ED symptoms. Higher scores in mental health and low perception of burden at baseline predicted improved caregiver QoL at 1-year follow-up. The caregivers' QoL improved if the perception of burden decreased over time and if depression of the patient improved. The results suggest that interventions directed to reduce the caregivers' perception of burden and to improve the mental health of the ED patients would lead to improved caregiver QoL. The high attrition rate at T2 made prediction analyses unviable.

Effectiveness of vaccination with 23-valent pneumococcal polysaccharide vaccine in preventing hospitalization with laboratory confirmed influenza during the 2009-2010 and 2010-2011 seasons.

BACKGROUND: Since influenza predisposes to bacterial pneumonia caused by Streptococcus pneumoniae, studies have suggested that pneumococcal vaccination might reduce its occurrence during pandemics. We assessed the effectiveness of pneumococcal polysaccharide vaccination alone and in combination with influenza vaccination in preventing influenza hospitalization during the 2009-2010 pandemic wave and 2010-2011 influenza epidemic. RESULTS: 1187 cases and 2328 controls were included. The adjusted estimate of effectiveness of pneumococcal vaccination in preventing influenza hospitalization was 41% (95% CI 8-62) in all patients and 43% (95% CI 2-78) in patients aged ≥ 65 y. The adjusted effectiveness of dual PPV23 and influenza vaccination was 81% (95% CI 65-90) in all patients and 76% (95% CI 46-90) in patients aged ≥ 65 y. The adjusted effectiveness of influenza vaccination alone was 58% (95% CI 38-72). METHODS: We conducted a multicenter case-control study in 36 Spanish hospitals. We selected patients aged ≥ 18 y hospitalized with confirmed influenza and two hospitalized controls per case, matched according to age, date of hospitalization and province of residence. Multivariate analysis was performed using conditional logistic regression. Subjects were considered vaccinated if they had received the pneumococcal or seasonal influenza vaccine>14 d (or>7 d for pandemic influenza vaccine) before the onset of symptoms (cases) or the onset of symptoms in matched cases (controls). CONCLUSIONS: In elderly people and adults with chronic illness, pneumococcal vaccination may reduce hospitalizations during the influenza season. In people vaccinated with both the influenza and pneumococcal vaccines, the benefit in hospitalizations avoided was greater than in those vaccinated only against influenza.

Social factors related to the clinical severity of influenza cases in Spain during the A (H1N1) 2009 virus pandemic.

BACKGROUND: During the 2009 influenza pandemic, a change in the type of patients most often affected by influenza was observed. The objective of this study was to assess the role of individual and social determinants in hospitalizations due to influenza A (H1N1) 2009 infection. METHODS: We studied hospitalized patients (cases) and outpatients (controls) with confirmed influenza A (H1N1) 2009 infection. A standardized questionnaire was used to collect data. Variables that might be related to the hospitalization of influenza cases were compared by estimation of the odds ratio (OR) and 95% confidence intervals (CI) and the variables entered into binomial logistic regression models. RESULTS: Hospitalization due to pandemic A (H1N1) 2009 influenza virus infections was associated with non-Caucasian ethnicity (OR: 2.18, 95% CI 1.17 - 4.08), overcrowding (OR: 2.84, 95% CI 1.20 - 6.72), comorbidity and the lack of previous preventive information (OR: 2.69, 95% CI: 1.50 - 4.83). Secondary or higher education was associated with a lower risk of hospitalization (OR 0.56, 95% CI: 0.36 - 0.87) CONCLUSIONS: In addition to individual factors such as comorbidity, other factors such as educational level, ethnicity or overcrowding were associated with hospitalization due to A (H1N1) 2009 influenza virus infections.

Potential for 2 priority systems to be used as appropriateness of indication tools for cataract surgery.

OBJECTIVE: To test 2 systems developed to prioritize patients on waiting lists for cataract surgery, the Western Canadian Waiting List (WCWL), and the Catalan Agency for Health Technology Assessment and Research Cataract Priority System (CCPS), as tools for judging the appropriateness of the intervention. DESIGN: Cross-sectional study with follow-up. PARTICIPANTS: A total of 1723 prospective patients awaiting cataract extraction in 5 hospitals were included in the study. METHODS: Priority scores using the WCWL and the CCPS systems were calculated and compared with visual acuity (VA) and visual function, measured by the Visual Function-14 index, before and after surgery. The sensitivity of each priority system to different groups was evaluated using a validated appropriateness of indications tool for cataract extraction. One-way analysis of variance and the Jonckheere-Terpstra test were used to detect differences in group scores. Spearman correlation coefficients were also examined. RESULTS: The WCWL and CCPS systems produced different mean values according to the preoperative levels of VA and visual function (p < 0.0001). Neither system showed a significant difference in mean scores based on gains in VA and visual function. Both systems produced different mean scores according to the different groups of appropriateness set by the appropriateness of indications tool for cataract (p < 0.0001). CONCLUSIONS: Given that appropriateness tools should correlate with outcomes, we recommend using neither the WCWL nor the CCPS for appropriateness purposes. Incorporating new criteria into these systems may improve the correlation with important outcomes and their adequacy for use as appropriateness tools.

Variability in the prioritization of patients for cataract extraction.

OBJECTIVE: The aim of this study was to compare time spent waiting for cataract extraction across various hospitals and to determine if it was influenced by severity of disease or patient characteristics. DESIGN: Ambispective cohort study. Clinical, sociodemographic, and health-related quality of life data were collected along with time spent on the waiting list. SETTING: Twelve hospitals in four regions of Spain. PARTICIPANTS: A total of 4043 patients who were waiting for being intervened on cataracts in the participant centres were recruited prospectively. Exposition variables Priority was assigned retrospectively using a previously validated scoring system. MAIN OUTCOME MEASURE: Time spent on waiting list. RESULTS: Statistically significant differences in time spent on the waiting list (P < 0.0001) were observed across the hospitals, even after adjusting for pathology, age, social dependency, laterality of the cataract and number of inappropriate interventions in each centre. Waiting time was not correlated with preintervention visual acuity, preintervention visual function measured by the VF-14 or priority score. CONCLUSIONS: Allocation of cataract surgery does not appear to be correlated with the need for surgery or the likely benefit to be derived from it in the Spanish participant centres. The use of explicit and standardized priority criteria could provide a fairer, more rational way to manage waiting lists for this procedure and may help to reduce unnecessary variation on access to health care.

Unmet needs and waiting list prioritization for knee arthroplasty.

BACKGROUND: There is a high volume of unmet needs for knee arthroplasty in the population despite the increase in surgery rates. Given the long waiting times to have a knee arthroplasty, some governments have proposed prioritization systems for patients on waiting lists based on their level of need. QUESTIONS/PURPOSES: We therefore estimated the needs and demand of knee arthroplasty in four regions of Spain during a 5-year period. METHODS: We developed a discrete event simulation model to reproduce the process of knee arthroplasty. The prioritization system was compared with the usual waiting list management strategy (by waiting time only). RESULTS: Under the prioritization system, patients saved an average of 4.5 months (95% confidence interval, 4.4-4.6 months) adjusted by level of need. The proportion of patients who experienced excessive waiting times was small and was associated with low levels of priority. The 5-year projection of the volume of unmet needs for knee arthroplasty remained stable; however, although the volume of need for the first knee arthroplasty decreased by 12%, the volume of need for an arthroplasty in the contralateral knee increased by 50%. CONCLUSIONS: The data suggested the prioritization system was more beneficial than assigning surgery by waiting time only. The 5-year projection of the volume of unmet needs for knee arthroplasty remained stable, despite the increase in the need for contralateral knee arthroplasty. LEVEL OF EVIDENCE: Level II, economic and decision analyses. See Guidelines for Authors for a complete description of levels of evidence.

Development of explicit criteria for total knee replacement.

OBJECTIVES: To develop and test an appropriateness of indications tool for total knee replacement (TKR) in patients with osteoarthritis. METHODS: Criteria were developed using a modified Delphi panel judgment. Another panel rated the same indications, and the results were compared with the main panel. Test-retest of the main panel was performed. Regression models were used to assess the contribution of each algorithm variable. A classification tree was developed. RESULTS: The procedure was considered appropriate in 167 (26.8%) scenarios, and there was agreement on 112 (67.1%) of them. When the rates of the main panel were compared with those of a second panel, the result was a kappa statistic of 0.75. The test-retest kappa for the main panel was 0.78. Neither in the first case nor in the second was there an instance in which a scenario classified as appropriate shifted to inappropriate or vice versa. The regression models showed that symptomatology and radiology were the variables that explained most of the variability of appropriateness as determined by panelists. In the classification tree performed, the probability of misclassification was 3.8% with 150 scenarios, of the 156 analyzed and classified correctly. CONCLUSIONS: The previous parameters tested showed acceptable results for an evaluation tool. These results support the use of this algorithm as an aid in formulating clinical practice guidelines and to promote the appropriateness of TKR.