Clinical relevance of an intervention assessed by a meta-analysis of randomized clinical trials.

OBJECTIVES: Many meta-analyses usually omit the number needed to treat, or perform the calculation incorrectly, despite its importance in clinical decision-making. Accordingly, we will explain in an easily understandable way how to perform this procedure to assess the clinical relevance of the intervention. STUDY DESIGN AND SETTING: The expressions of the Cochrane Library and the concepts of clinical relevance and evidence-based medicine were applied. Simple cutoff points were also established to facilitate the task of interpreting results. The method was applied to two published meta-analyses to illustrate its application to real cases (treatment nonadherence). RESULTS: In the first example, with a risk in the control group ranging from 0.22 to 0.70, sending mobile phone messages to remind chronic patients to take their medication is clinically relevant with a high degree of evidence. For the second example (single-pill regimen in patients suffering from hypertension and/or dyslipidemia after 6 months), the range of the assumed control risk was between 0.28 and 0.57. CONCLUSION: The constructed algorithm could be applied to published meta-analyses or incorporated systematically in all meta-analyses with these characteristics.

Development, and Internal, and External Validation of a Scoring System to Predict 30-Day Mortality after Having a Traffic Accident Traveling by Private Car or Van: An Analysis of 164,790 Subjects and 79,664 Accidents.

Predictive factors for fatal traffic accidents have been determined, but not addressed collectively through a predictive model to help determine the probability of mortality and thereby ascertain key points for intervening and decreasing that probability. Data on all road traffic accidents with victims involving a private car or van occurring in Spain in 2015 (164,790 subjects and 79,664 accidents) were analyzed, evaluating 30-day mortality following the accident. As candidate predictors of mortality, variables associated with the accident (weekend, time, number of vehicles, road, brightness, and weather) associated with the vehicle (type and age of vehicle, and other types of vehicles in the accident) and associated with individuals (gender, age, seat belt, and position in the vehicle) were examined. The sample was divided into two groups. In one group, a logistic regression model adapted to a points system was constructed and internally validated, and in the other group the model was externally validated. The points system obtained good discrimination and calibration in both the internal and the external validation. Consequently, a simple tool is available to determine the risk of mortality following a traffic accident, which could be validated in other countries.

A method to validate scoring systems based on logistic regression models to predict binary outcomes via a mobile application for Android with an example of a real case.

BACKGROUND AND OBJECTIVES: To use a points system based on a logistic regression model to predict a binary event in a given population, the validation of this system is necessary. The most correct way to do this is to calculate discrimination and calibration using bootstrapping. Discrimination can be addressed through the area under the receiver operating characteristic curve (AUC) and calibration through the representation of the smoothed calibration plot (most recommended method). As this is not a simple task, we developed a methodology to construct a mobile application in Android to perform this task. METHODS: The construction of the application is based on source code written in language supported by Android. It is designed to use a database of subjects to be analyzed and to be able to apply statistical methods widely used in the scientific literature to validate a points system (bootstrap, AUC, logistic regression models and smooth curves). As an example our methodology was applied on simulated points system data (doi: 10.1111/ijcp.12851) to predict mortality on admission to intensive care units (Google Play: ICU mortality). The results were compared with those obtained applying the same methods in the R statistical package. RESULTS: No differences were found between the results obtained in the mobile application and those from the R statistical package, an expected result when applying the same mathematical techniques. CONCLUSIONS: Our methodology may be applied to other point systems for predicting binary events, as well as to other types of predictive models.

A one-year risk score to predict all-cause mortality in hypertensive inpatients.

The aim of this study was to construct and internally validate a scoring system to estimate the probability of death in hypertensive inpatients. Existing predictive models do not meet all the indications for clinical application because they were constructed in patients enrolled in clinical trials and did not use the recommended statistical methodology. This cohort study comprised 302 hypertensive patients hospitalized between 2015 and 2017 in Spain. The main variable was time-to-death (all-cause mortality). Secondary variables (potential predictors of the model) were: age, gender, smoking, blood pressure, Charlson Comorbidity Index (CCI), physical activity, diet and quality of life. A Cox model was constructed and adapted to a points system to predict mortality one year from admission. The model was internally validated by bootstrapping, assessing both discrimination and calibration. The system was integrated into a mobile application for Android. During the study, 63 patients died (20.9%). The points system prognostic variables were: gender, CCI, personal care and daily activities. Internal validation showed good discrimination (mean C statistic of 0.76) and calibration (observed probabilities adjusted to predicted probabilities). In conclusion, a points system was developed to determine the one-year mortality risk for hypertensive inpatients. This system is very simple to use and has been internally validated. Clinically, we could monitor more closely those patients with a higher risk of mortality to improve their prognosis and quality of life. However, the system must be externally validated to be applied in other geographic areas.

A scoring system to predict recurrence in breast cancer patients.

OBJECTIVE: Current breast cancer recurrence prediction models have limitations for clinical practice (statistical methodology, simplicity and specific populations). We therefore developed a new model that overcomes these limitations. METHODS: This cohort study comprised 272 patients with breast cancer followed between 2003 and 2016. The main variable was time-to-recurrence (locoregional and/or metastasis) and secondary variables were its risk factors: age, postmenopause, grade, oestrogen receptor, progesterone receptor, c-erbB2 status, stage, multicentricity, diagnosis and treatment. A Cox model to predict recurrence was estimated with the secondary variables, and this was adapted to a points system to predict risk at 5 and 10 years from diagnosis. The model was validated internally by bootstrapping, calculating the C statistic and smooth calibration (splines). The system was integrated into a mobile application for Android. RESULTS: Of the 272 patients with breast cancer, 47 (17.3%) developed recurrence in a mean time of 8.6 ±â€¯3.5 years. The system variables were: age, grade, multicentricity and stage. Validation by bootstrapping showed good discrimination and calibration. CONCLUSIONS: A points system has been developed to predict breast cancer recurrence at 5 and 10 years.

A scoring system to detect fixed airflow limitation in smokers from simple easy-to-use parameters.

No validated screening method currently exists for Chronic Obstructive Pulmonary Disease (COPD) in smokers. Therefore, we constructed a predictive model with simple parameters that can be applied for COPD screening to detect fixed airflow limitation. This observational cross-sectional study included a random sample of 222 smokers with no previous diagnosis of COPD undertaken in a Spanish region in 2014-2016. The main variable was fixed airflow limitation by spirometry. The secondary variables (COPD factors) were: age, gender, smoking (pack-years and Fagerström test), body mass index, educational level, respiratory symptoms and exacerbations. A points system was developed to predict fixed airflow limitation based on secondary variables. The model was validated internally through bootstrapping, determining discrimination and calibration. The system was then integrated into a mobile application for Android. Fifty-seven patients (25.7%) presented fixed airflow limitation. The points system included as predictors: age, pack-years, Fagerström test and presence of respiratory symptoms. Internal validation of the system was very satisfactory, both in discrimination and calibration. In conclusion, a points system has been constructed to predict fixed airflow limitation in smokers with no previous COPD. This system can be integrated as a screening tool, though it should be externally validated in other geographical regions.

A scoring system to predict 5-year mortality in patients diagnosed with laryngeal glottic cancer.

Only one prognostic model for laryngeal cancer has been published, but it has not been properly validated and is only applicable to patients treated with radiotherapy. Consequently, we constructed, internally validated and implemented in an App (Android), a predictive model of 5-year mortality in patients with glottic cancer in a cohort study of 189 patients with glottic cancer in 2004-2016 in Spain. The main variable was time-to-death. Secondary variables were age, gender, TNM, stage, smoking, alcohol consumption, histology and treatment. A scoring system to predict mortality at 5 years was constructed, validated internally by bootstrapping and then integrated into an Android app. In all, 70 patients died (37.0%, 76 deaths per 1,000 patient-years). The predictive model had the following prognostic factors: larger tumour size, greater degree of lymph node metastasis, higher stage, smoking and alcohol consumption. The internal validation of the model through bootstrapping was satisfactory. In conclusion, a points system to predict mortality at 5 years in patients with glottic cancer has been constructed, internally validated and integrated into an Android application. External validation is suggested to make available a quick and simple tool to establish the prognosis for these patients.

A predictive model for recurrence in patients with glottic cancer implemented in a mobile application for Android.

OBJECTIVES: The existing predictive models of laryngeal cancer recurrence present limitations for clinical practice. Therefore, we constructed, internally validated and implemented in a mobile application (Android) a new model based on a points system taking into account the internationally recommended statistical methodology. MATERIALS AND METHODS: This longitudinal prospective study included 189 patients with glottic cancer in 2004-2016 in a Spanish region. The main variable was time-to-recurrence, and its potential predictors were: age, gender, TNM classification, stage, smoking, alcohol consumption, and histology. A points system was developed to predict five-year risk of recurrence based on a Cox model. This was validated internally by bootstrapping, determining discrimination (C-statistics) and calibration (smooth curves). RESULTS: A total of 77 patients presented recurrence (40.7%) in a mean follow-up period of 3.4 ±â€¯3.0 years. The factors in the model were: age, lymph node stage, alcohol consumption and stage. Discrimination and calibration were satisfactory. CONCLUSION: A points system was developed to obtain the probability of recurrence of laryngeal glottic cancer in five years, using five clinical variables. Our system should be validated externally in other geographical areas.

Construction, internal validation and implementation in a mobile application of a scoring system to predict nonadherence to proton pump inhibitors.

BACKGROUND: Other studies have assessed nonadherence to proton pump inhibitors (PPIs), but none has developed a screening test for its detection. OBJECTIVES: To construct and internally validate a predictive model for nonadherence to PPIs. METHODS: This prospective observational study with a one-month follow-up was carried out in 2013 in Spain, and included 302 patients with a prescription for PPIs. The primary variable was nonadherence to PPIs (pill count). Secondary variables were gender, age, antidepressants, type of PPI, non-guideline-recommended prescription (NGRP) of PPIs, and total number of drugs. With the secondary variables, a binary logistic regression model to predict nonadherence was constructed and adapted to a points system. The ROC curve, with its area (AUC), was calculated and the optimal cut-off point was established. The points system was internally validated through 1,000 bootstrap samples and implemented in a mobile application (Android). RESULTS: The points system had three prognostic variables: total number of drugs, NGRP of PPIs, and antidepressants. The AUC was 0.87 (95% CI [0.83-0.91], p < 0.001). The test yielded a sensitivity of 0.80 (95% CI [0.70-0.87]) and a specificity of 0.82 (95% CI [0.76-0.87]). The three parameters were very similar in the bootstrap validation. CONCLUSIONS: A points system to predict nonadherence to PPIs has been constructed, internally validated and implemented in a mobile application. Provided similar results are obtained in external validation studies, we will have a screening tool to detect nonadherence to PPIs.

Sample size calculation to externally validate scoring systems based on logistic regression models.

BACKGROUND: A sample size containing at least 100 events and 100 non-events has been suggested to validate a predictive model, regardless of the model being validated and that certain factors can influence calibration of the predictive model (discrimination, parameterization and incidence). Scoring systems based on binary logistic regression models are a specific type of predictive model. OBJECTIVE: The aim of this study was to develop an algorithm to determine the sample size for validating a scoring system based on a binary logistic regression model and to apply it to a case study. METHODS: The algorithm was based on bootstrap samples in which the area under the ROC curve, the observed event probabilities through smooth curves, and a measure to determine the lack of calibration (estimated calibration index) were calculated. To illustrate its use for interested researchers, the algorithm was applied to a scoring system, based on a binary logistic regression model, to determine mortality in intensive care units. RESULTS: In the case study provided, the algorithm obtained a sample size with 69 events, which is lower than the value suggested in the literature. CONCLUSION: An algorithm is provided for finding the appropriate sample size to validate scoring systems based on binary logistic regression models. This could be applied to determine the sample size in other similar cases.

A scoring system to predict breast cancer mortality at 5 and 10 years.

Although predictive models exist for mortality in breast cancer (BC) (generally all cause-mortality), they are not applicable to all patients and their statistical methodology is not the most powerful to develop a predictive model. Consequently, we developed a predictive model specific for BC mortality at 5 and 10 years resolving the above issues. This cohort study included 287 patients diagnosed with BC in a Spanish region in 2003-2016. MAIN OUTCOME VARIABLE: time-to-BC death. Secondary variables: age, personal history of breast surgery, personal history of any cancer/BC, premenopause, postmenopause, grade, estrogen receptor, progesterone receptor, c-erbB2, TNM stage, multicentricity/multifocality, diagnosis and treatment. A points system was constructed to predict BC mortality at 5 and 10 years. The model was internally validated by bootstrapping. The points system was integrated into a mobile application for Android. Mean follow-up was 8.6 ± 3.5 years and 55 patients died of BC. The points system included age, personal history of BC, grade, TNM stage and multicentricity. Validation was satisfactory, in both discrimination and calibration. In conclusion, we constructed and internally validated a scoring system for predicting BC mortality at 5 and 10 years. External validation studies are needed for its use in other geographical areas.

Non-guideline-recommended prescribing of proton pump inhibitors in the general population.

OBJECTIVES: To determine the magnitude of non-guideline-recommended prescribing (NGRP) of proton pump inhibitors (PPIs) in the general population, its associated factors and expense. METHODS: We undertook a cross-sectional observational study in three community pharmacies in a Spanish region in 2013 involving a total of 302 patients with a prescription for PPIs. The main variable was the NGRP of PPIs. Secondary variables were: gender, age, antidepressants, osteoporosis, osteoarthritis, prescription cost per month and total number of chronic diseases. The cost associated with NGRP was calculated. To evaluate the associated factors, a multivariate binary logistic regression model was constructed and the adjusted odds ratios (OR) were obtained. RESULTS: NGRP was observed in 192 cases (63.6%). The average cost associated with NGRP per prescription was 3.24 euros per month. The factors significantly associated with NGRP (p < .05) were: antidepressants (OR = 2.66, p = .001), osteoporosis (OR = 3.53, p = .001), osteoarthritis (OR = 3.57, p < .001) and number of chronic diseases (OR = 0.73, p = .003). CONCLUSION: A novel approach was used to quantify the NGRP of PPIs in a Spanish community, as well as the associated economic costs. Qualitative studies are needed to better understand the causes of NGRP of PPIs. This analysis will aid in designing interventions to minimize this problem. LIMITATIONS: Qualitative studies are needed to better understand the attitude of health professionals when prescribing PPIs.

A predictive screening tool to detect diabetic retinopathy or macular edema in primary health care: construction, validation and implementation on a mobile application.

The most described techniques used to detect diabetic retinopathy and diabetic macular edema have to be interpreted correctly, such that a person not specialized in ophthalmology, as is usually the case of a primary care physician, may experience difficulties with their interpretation; therefore we constructed, validated and implemented as a mobile app a new tool to detect diabetic retinopathy or diabetic macular edema (DRDME) using simple objective variables. We undertook a cross-sectional, observational study of a sample of 142 eyes from Spanish diabetic patients suspected of having DRDME in 2012-2013. Our outcome was DRDME and the secondary variables were: type of diabetes, gender, age, glycated hemoglobin (HbA1c), foveal thickness and visual acuity (best corrected). The sample was divided into two parts: 80% to construct the tool and 20% to validate it. A binary logistic regression model was used to predict DRDME. The resulting model was transformed into a scoring system. The area under the ROC curve (AUC) was calculated and risk groups established. The tool was validated by calculating the AUC and comparing expected events with observed events. The construction sample (n = 106) had 35 DRDME (95% CI [24.1-42.0]), and the validation sample (n = 36) had 12 DRDME (95% CI [17.9-48.7]). Factors associated with DRDME were: HbA1c (per 1%) (OR = 1.36, 95% CI [0.93-1.98], p = 0.113), foveal thickness (per 1 µm) (OR = 1.03, 95% CI [1.01-1.04], p < 0.001) and visual acuity (per unit) (OR = 0.14, 95% CI [0.00-0.16], p < 0.001). AUC for the validation: 0.90 (95% CI [0.75-1.00], p < 0.001). No significant differences were found between the expected and the observed outcomes (p = 0.422). In conclusion, we constructed and validated a simple rapid tool to determine whether a diabetic patient suspected of having DRDME really has it. This tool has been implemented on a mobile app. Further validation studies are required in the general diabetic population.

Scoring System for Mortality in Patients Diagnosed with and Treated Surgically for Differentiated Thyroid Carcinoma with a 20-Year Follow-Up.

BACKGROUND: Differentiated thyroid carcinoma (DTC) is associated with an increased mortality. Few studies have constructed predictive models of all-cause mortality with a high discriminating power for patients with this disease that would enable us to determine which patients are more likely to die. OBJECTIVE: To construct a predictive model of all-cause mortality at 5, 10, 15 and 20 years for patients diagnosed with and treated surgically for DTC for use as a mobile application. DESIGN: We undertook a retrospective cohort study using data from 1984 to 2013. SETTING: All patients diagnosed with and treated surgically for DTC at a general university hospital covering a population of around 200,000 inhabitants in Spain. PARTICIPANTS: The study involved 201 patients diagnosed with and treated surgically for DTC (174, papillary; 27, follicular). EXPOSURES: Age, gender, town, family history, type of surgery, type of cancer, histological subtype, microcarcinoma, multicentricity, TNM staging system, diagnostic stage, permanent post-operative complications, local and regional tumor persistence, distant metastasis, and radioiodine therapy. MAIN OUTCOME MEASURE: All-cause mortality. METHODS: A Cox multivariate regression model was constructed to determine which variables at diagnosis were associated with mortality. Using the model a risk table was constructed based on the sum of all points to estimate the likelihood of death. This was then incorporated into a mobile application. RESULTS: The mean follow-up was 8.8±6.7 years. All-cause mortality was 12.9% (95% confidence interval [CI]: 8.3-17.6%). Predictive variables: older age, local tumor persistence and distant metastasis. The area under the ROC curve was 0.81 (95% CI: 0.72-0.91, p<0.001). CONCLUSION: This study provides a practical clinical tool giving a simple and rapid indication (via a mobile application) of which patients with DTC are at risk of dying in 5, 10, 15 or 20 years. Nonetheless, caution should be exercised until validation studies have corroborated our results.

A four-year cardiovascular risk score for type 2 diabetic inpatients.

As cardiovascular risk tables currently in use were constructed using data from the general population, the cardiovascular risk of patients admitted via the hospital emergency department may be underestimated. Accordingly, we constructed a predictive model for the appearance of cardiovascular diseases in patients with type 2 diabetes admitted via the emergency department. We undertook a four-year follow-up of a cohort of 112 adult patients with type 2 diabetes admitted via the emergency department for any cause except patients admitted with acute myocardial infarction, stroke, cancer, or a palliative status. The sample was selected randomly between 2010 and 2012. The primary outcome was time to cardiovascular disease. Other variables (at baseline) were gender, age, heart failure, renal failure, depression, asthma/chronic obstructive pulmonary disease, hypertension, dyslipidaemia, insulin, smoking, admission for cardiovascular causes, pills per day, walking habit, fasting blood glucose and creatinine. A cardiovascular risk table was constructed based on the score to estimate the likelihood of cardiovascular disease. Risk groups were established and the c-statistic was calculated. Over a mean follow-up of 2.31 years, 39 patients had cardiovascular disease (34.8%, 95% CI [26.0-43.6%]). Predictive factors were gender, age, hypertension, renal failure, insulin, admission due to cardiovascular reasons and walking habit. The c-statistic was 0.734 (standard error: 0.049). After validation, this study will provide a tool for the primary health care services to enable the short-term prediction of cardiovascular disease after hospital discharge in patients with type 2 diabetes admitted via the emergency department.

Construction and validation of a model to predict nonadherence to guidelines for prescribing antiplatelet therapy to hypertensive patients.

OBJECTIVE: To construct and validate a model to predict nonadherence to guidelines for prescribing antiplatelet therapy (NGAT) to hypertensive patients. METHODS: This 3 month prospective study was undertaken in 2007-2009 to determine whether 712 hypertensive patients were or were not being prescribed antiplatelet therapy. OUTCOME: NGAT according to clinical guidelines (just for patients in secondary prevention or with Systematic COronary Risk Evaluation (SCORE) ≥10%). Secondary variables: Duration of hypertension (years), blood pressure (BP), age, gender, smoking, diabetes, dyslipidemia, cardiovascular disease, lipid parameters, SCORE. Of the whole sample 80% was used to construct the model and 20% to validate it. To construct the model, we performed a multivariate logistic regression model which was adapted to be a scoring system with risk groups. The adjusted odds ratios (ORs) were obtained through the model. To validate the model we calculated the area under the ROC curve (AUC) and then compared the expected and the observed NGAT. The final model was adapted for use as a mobile application. RESULTS: NGAT: 18.5%, construction; 17.9%, validation. FACTORS: higher duration of hypertension diagnosis, higher systolic BP, older age, male gender, smoking, diabetes, dyslipidemia and cardiovascular disease. VALIDATION: AUC = 0.82 (95% CI: 0.74-0.90, p < 0.001), with no differences between the observed and the expected NGAT (p = 0.334). CONCLUSION: A tool was constructed and validated to predict NGAT. The associated factors were related with a greater cardiovascular risk. The scoring system has to be validated in other areas.