Decision Threshold Research Articles

Development of a postoperative nomogram to predict disease-specific mortality in gastric cancer: A competing risk analysis using variables the AJCC recommends be collected and registered.

e16099 Background: The American Joint Committee on Cancer (AJCC) has increasingly recognized the need for individual risk prediction model for the era of tailored therapy. In several reports, gastric cancer postoperative nomograms for predicting overall or disease-specific survival have been described. The AJCC Precision Medicine Core introduced the attractiveness of disease-specific mortality (DSM) as an endpoint of risk model. This study aimed to develop the first pretreatment gastric cancer postoperative nomogram for predicting DSM that considers competing risks (CRs). Methods: The nomogram was developed using data of 2,501 patients with primary gastric cancer who are potential candidate for adjuvant chemotherapy based on NCCN guideline, and it was created with a multivariable Fine and Gray subdistribution proportional hazard regression model. Fifteen pathological or non-tumor-related variables (age, sex tumor location, tumor size, macroscopic type, histology, depth (pT), number of positive nodes (pN), number of negative nodes, location of positive nodes, lymphovascular invasion, lavage cytology (CY), tumor margin, serum CEA and serum CA19-9) were collected. They are recommended to collect and register by the AJCC. The model was validated internally using measures of discrimination (Harrell’s C-index), calibration and decision curve analysis. Results: In the development procedure, multivariable analysis for DSM selected 12 variables (age, tumor location, macroscopic type, histology, pT, pN, number of negative nodes, location of positive nodes, lymphovascular invasion, CY, tumor margin and serum CEA) for constructing the nomogram. The developed nomogram showed good discrimination, with a C-index of 0.802(0.786-0.817) that of the American Joint Committee on Cancer (AJCC) pathological stage was 0.714 (0.697-0.734). The calibration appeared to be accurate for the 5-year prediction. Compared with scenarios in which no prediction model was used for postoperative treatment decision making (i.e., assume all or assume none), the nomograms had a favorable net benefit across a wide range of decision threshold probabilities between 5-year DSM probabilities of approximately 5 and 95%. Conclusions: This new postoperative risk model accurately predicts DSM in gastric cancer and can be used for patient counseling in clinical practice.

Review of the clinical outcomes of therapeutic bronchoscopy for central airway obstruction.

Central airway obstruction (CAO) is a debilitating condition with a significant impact on patient's quality of life and risk of hospitalization from respiratory failure. The causes of CAO can be both benign and malignant. Benign CAO may be idiopathic or secondary to other disease processes (infection, intubation, tracheostomy, etc.). Malignant central airway obstruction (MCAO) may occur in patients with primary lung malignancy as well as metastasis from other malignancies including renal cell, colon, and breast. In a cohort review, MCAO was found in up to 13% of patients with newly diagnosed lung cancer. The obstruction may occur either due to endoluminal disease, extrinsic compression, or a combination of both. Several bronchoscopic tools are available to manage such obstruction. Practice patterns and tools used to relieve CAO vary between institutions and may depend on physician preference, patient characteristics, emergency nature of the procedure, and nature of the obstruction. To quantify the effect and added value of such interventions, it is crucial to understand the clinical impact these interventions have on patients. The clinical impact of therapeutic bronchoscopy (TB) must then be weighed against the potential complications to justify its value. Early studies of TB for CAO included patients with both malignant and benign etiologies. The study population's heterogeneity makes it difficult to determine how TB affects clinical outcomes, as clinical outcomes are disease specific. The impact of TB for a MCAO may be different when compared to a benign CAO. Similarly, the clinical outcome of treating an idiopathic benign CAO may be different than that of a post tracheostomy airway obstruction. In this article, we will focus on the clinical outcomes of TB in MCAO. TB has been shown to have a clear impact on weaning from mechanical ventilation, dyspnea, health-related quality of life, survival and quality adjusted survival. The potential impact of TB on these outcomes should be weighed against the potential risk of complications. Understanding the factors associated with improved clinical outcomes will help physicians decide when and if TB is helpful. Future studies should focus on creating a decision analysis tool to further define decision thresholds.

Validation of the ADFICE_IT Models for Predicting Falls and Recurrent Falls in Geriatric Outpatients

ObjectivesBefore being used in clinical practice, a prediction model should be tested in patients whose data were not used in model development. Previously, we developed the ADFICE_IT models for predicting any fall and recurrent falls, referred as Any_fall and Recur_fall. In this study, we externally validated the models and compared their clinical value to a practical screening strategy where patients are screened for falls history alone. DesignRetrospective, combined analysis of 2 prospective cohorts. Setting and ParticipantsData were included of 1125 patients (aged ≥65 years) who visited the geriatrics department or the emergency department. MethodsWe evaluated the models' discrimination using the C-statistic. Models were updated using logistic regression if calibration intercept or slope values deviated significantly from their ideal values. Decision curve analysis was applied to compare the models’ clinical value (ie, net benefit) against that of falls history for different decision thresholds. ResultsDuring the 1-year follow-up, 428 participants (42.7%) endured 1 or more falls, and 224 participants (23.1%) endured a recurrent fall (≥2 falls). C-statistic values were 0.66 (95% CI 0.63-0.69) and 0.69 (95% CI 0.65-0.72) for the Any_fall and Recur_fall models, respectively. Any_fall overestimated the fall risk and we therefore updated only its intercept whereas Recur_fall showed good calibration and required no update. Compared with falls history, Any_fall and Recur_fall showed greater net benefit for decision thresholds of 35% to 60% and 15% to 45%, respectively. Conclusions and ImplicationsThe models performed similarly in this data set of geriatric outpatients as in the development sample. This suggests that fall-risk assessment tools that were developed in community-dwelling older adults may perform well in geriatric outpatients. We found that in geriatric outpatients the models have greater clinical value across a wide range of decision thresholds compared with screening for falls history alone.

Kernel similarity-based multigranulation three-way decision approach to hypertension risk assessment with multi-source and multi-level structure data

Early intervention and prevention of chronic diseases is a fundamental way to reduce disease incidence. Due to the complex pathogenic factors of disease, the assessment of disease risk requires consideration of the influence of multiple factors. In clinical practice, disease risk assessment may face data with multi-source, multi-type and multiple indicators of examination items. Mining useful information and rules in this complex information is the foundation for further decisions. However, the existing knowledge representation methods cannot comprehensively and accurately describe these characteristic of data. Therefore, we introduce the concept of multiple hybrid attribute information systems (MHAISs). Then, we defined the binary relations over MHAIS by fusing different types of kernel functions. On this basis, we construct a variable precision multigranulation kernel rough set (VPMGKRS), and propose a multigranulation three-way decision method over MHAIS. In addition, considering the individual difference between decision objects and the diversity of data characteristic among sources, we calculate the loss functions under different granular with the help of conditional probabilities, and then obtain the thresholds for the three-way decisions. Finally, we use 712 clinical random samples to conduct a simulation analysis applying the proposed method to assess the risk of hypertension. The experiment result confirmed the applicability and validity of the model.

W-band RoF polarization multiplexing system supports 1048576 QAM with delta-sigma modulation.

We propose and experimentally verify a photonics-aided W-band millimeter wave (MMW) radio-over-fiber (RoF) polarization-multiplexed envelope detection system for high-order quadrature amplitude modulation (QAM) signals. To solve the problem of low spectral efficiency of common public radio interface (CPRI) and severe distortion of high-order QAM of envelope detection, quantization noise suppressed delta-sigma modulation (DSM) is introduced into the system. The experimental results show that the system can transmit 131072 QAM signals when meeting the error vector magnitude (EVM) requirements of 5G new radio (NR), and transmit 1048576 QAM signals when meeting the soft decision threshold (SD@20%).

Osteoporosis: Spotlight on current approaches to pharmacological treatment.

Despite the availability of safe and effective anti-osteoporosis treatments, osteoporosis continues to be undertreated. The increase in fragility fractures, which is the main clinical consequence of osteoporosis, is a major problem for healthcare systems of countries. A broad range of drugs including antiresorptive and anabolic agents are used in the pharmacological treatment of osteoporosis. Fracture risk assessment in drug selection is of utmost importance in terms of guiding treatment. The recommended thresholds for osteoporosis treatment decision making are based on major osteoporotic and hip fracture probabilities from the Fracture Risk Assessment Tool (FRAX®). Currently, antiresorptive agents are usually the first choice to increase bone mineral density (BMD) and reduce the fracture risk. Bisphosphonates and antiresorptive drugs such as denosumab, a nuclear factor kappa-B ligand (RANKL) inhibitor, are the most widely used drugs in the treatment of osteoporosis. Bisphosphonates alone are unlikely to provide long-term protection against fracture and restore BMD in patients with severe osteoporosis and high fracture risk. In such patients, treatment with an anabolic agent such as teriparatide, abaloparatide, or romosozumab should be ideally initiated to achieve maximal gain in bone mass and preserve the microarchitecture. Ideally, an antiresorptive drug should be continued to maintain gain in bone mass.

Steganalysis of Neural Networks Based on Symmetric Histogram Distribution

Deep neural networks have achieved remarkable success in various fields of artificial intelligence. However, these models, which contain a large number of parameters, are widely distributed and disseminated by researchers, engineers, and even unauthorized users. Except for intelligent tasks, typically overparameterized deep neural networks have become new digital covers for data hiding, which may pose significant security challenges to AI systems. To address this issue, this paper proposes a symmetric steganalysis scheme specifically designed for neural networks trained for image classification tasks. The proposed method focuses on detecting the presence of additional data without access to the internal structure or parameters of the host network. It employs a well-designed method based on histogram distribution to find the optimal decision threshold, with a symmetric determining rule where the original networks and stego networks undergo two highly symmetrical flows to generate the classification labels; the method has been shown to be practical and effective. SVM and ensemble classifiers were chosen as the binary classifier for their applicability to feature vectors output from neural networks based on different datasets and network structures. This scheme is the first of its kind, focusing on steganalysis for neural networks based on the distribution of network output, compared to conventional digital media such as images, audio, and video. Overall, the proposed scheme offers a promising approach to enhancing the security of deep neural networks against data hiding attacks.

From the Editor and Chair of the Award Committee: 2022 Clemen–KleinmuntzDecision AnalysisBest Paper Award

From the Editor and Chair of the Award Committee: 2022 Clemen–Kleinmuntz<i>Decision Analysis</i>Best Paper Award

De-identified Bayesian personal identity matching for privacy-preserving record linkage despite errors: development and validation

BackgroundEpidemiological research may require linkage of information from multiple organizations. This can bring two problems: (1) the information governance desirability of linkage without sharing direct identifiers, and (2) a requirement to link databases without a common person-unique identifier.MethodsWe develop a Bayesian matching technique to solve both. We provide an open-source software implementation capable of de-identified probabilistic matching despite discrepancies, via fuzzy representations and complete mismatches, plus de-identified deterministic matching if required. We validate the technique by testing linkage between multiple medical records systems in a UK National Health Service Trust, examining the effects of decision thresholds on linkage accuracy. We report demographic factors associated with correct linkage.ResultsThe system supports dates of birth (DOBs), forenames, surnames, three-state gender, and UK postcodes. Fuzzy representations are supported for all except gender, and there is support for additional transformations, such as accent misrepresentation, variation for multi-part surnames, and name re-ordering. Calculated log odds predicted a proband’s presence in the sample database with an area under the receiver operating curve of 0.997–0.999 for non-self database comparisons. Log odds were converted to a decision via a consideration threshold θ and a leader advantage threshold δ. Defaults were chosen to penalize misidentification 20-fold versus linkage failure. By default, complete DOB mismatches were disallowed for computational efficiency. At these settings, for non-self database comparisons, the mean probability of a proband being correctly declared to be in the sample was 0.965 (range 0.931–0.994), and the misidentification rate was 0.00249 (range 0.00123–0.00429). Correct linkage was positively associated with male gender, Black or mixed ethnicity, and the presence of diagnostic codes for severe mental illnesses or other mental disorders, and negatively associated with birth year, unknown ethnicity, residential area deprivation, and presence of a pseudopostcode (e.g. indicating homelessness). Accuracy rates would be improved further if person-unique identifiers were also used, as supported by the software. Our two largest databases were linked in 44 min via an interpreted programming language.ConclusionsFully de-identified matching with high accuracy is feasible without a person-unique identifier and appropriate software is freely available.

Performance of cascaded gain saturated and fixed gain optical amplifier FSO communication systems limited by scintillation and pointing error

A major limitation to achieving acceptable system performance in free space optical (FSO) communication systems is the presence of atmospheric turbulence (AT), which results in received signal power fluctuations. In addition to the signal power fluctuations at the receiving end, pointing errors (PEs) can also significantly impair system performance. While achieving reliable communication over long distances is desirable, it is challenging because the AT effects and PEs are more pronounced at longer communication distances. Using Monte Carlo simulation methods, the impact of amplified spontaneous emission noise, AT, PE and geometric spread on cascaded gain saturated and fixed gain optical amplifier (OA) FSO communication systems is investigated in this paper. Results are obtained for different AT regimes, normalised PE standard deviations and normalised beam widths employing non-adaptive and adaptive decision thresholding schemes at the receiver. Results obtained show that increasing the size of the receiver aperture does not have any impact on the average bit error rate (BER) when the PEs are minimal. Also, the BER performances obtained with systems using a smaller receiver aperture are better than those obtained with systems using a larger receiver aperture when PEs are significant. Results obtained also show that even though unacceptable BER performances were obtained under moderate and strong atmospheric regimes when the decision threshold of the receiver is non-adaptive, the advantage of using gain saturated OAs is still evident as gain saturated OAs performed better than fixed gain OAs when the decision threshold of the receiver is non-adaptive.

Ecological forecasting models: Accuracy versus decisional quality

We consider here forecasting models in ecology or in agronomy, aiming at decision making based upon exceeding a quantitative threshold. We address specifically how to link the intrinsic quality of the model (its accuracy) with its decisional quality, ie its capacity to avoid false decisions and their associated costs. The accuracy of the model can be evaluated by the ρ of the regression of observed values versus estimated ones or by the determination coefficient R2. We show that the decisional quality depends not only of this accuracy but also of the threshold retained to make the decision as well as on the state of nature. The two kinds of decisional errors consists either in deciding no action while an action is required (false negatives) or to act while it is useless (false positives). We also prove that the costs associated to those decisions depend also both of the accuracy of the model and of the value of the decision threshold.

Determinants of referral for suspected coronary artery disease: a qualitative study based on decision thresholds

BackgroundChest pain is a frequent consultation issue in primary care, with coronary artery disease (CAD) being a serious potential cause. Primary care physicians (PCPs) assess the probability for CAD and refer patients to secondary care if necessary. Our aim was to explore PCPs’ referral decisions, and to investigate determinants which influenced those decisions.MethodsPCPs working in Hesse, Germany, were interviewed in a qualitative study. We used ‘stimulated recall’ with participants to discuss patients with suspected CAD. With a sample size of 26 cases from nine practices we reached inductive thematic saturation. Interviews were audio-recorded, transcribed verbatim and analyzed by inductive-deductive thematic content analysis. For the final interpretation of the material, we used the concept of decision thresholds proposed by Pauker and Kassirer.ResultsPCPs reflected on their decisions for or against a referral. Aside from patient characteristics determining disease probability, we identified general factors which can be understood as influencing the referral threshold. These factors relate to the practice environment, to PCPs themselves and to non-diagnostic patient characteristics. Proximity of specialist practice, relationship with specialist colleagues, and trust played a role. PCPs sometimes felt that invasive procedures were performed too easily. They tried to steer their patients through the system with the intent to avoid over-treatment. Most PCPs were unaware of guidelines but relied on informal local consensus, largely influenced by specialists. As a result, PCPs gatekeeping role was limited.ConclusionsWe could identify a large number of factors that impact referral for suspected CAD. Several of these factors offer possibilities to improve care at the clinical and system level. The threshold model proposed by Pauker and Kassirer was a useful framework for this kind of data analysis.

Diagnostic performance of deep learning in infectious keratitis: a systematic review and meta-analysis protocol

IntroductionInfectious keratitis (IK) represents the fifth-leading cause of blindness worldwide. A delay in diagnosis is often a major factor in progression to irreversible visual impairment and/or blindness from IK. The...

The impact of threshold decision mechanisms of collective behavior on disease spread

Humans are a hyper-social species, which greatly impacts the spread of infectious diseases. How do social dynamics impact epidemiology and what are the implications for public health policy? Here, we develop a model of disease transmission that incorporates social dynamics and a behavior that reduces the spread of disease, a voluntary nonpharmaceutical intervention (NPI). We use a "tipping-point" dynamic, previously used in the sociological literature, where individuals adopt a behavior given a sufficient prevalence of the behavior in the population. The thresholds at which individuals adopt the NPI behavior are modulated by the perceived risk of infection, i.e., the disease prevalence and transmission rate, costs to adopt the NPI behavior, and the behavior of others. Social conformity creates a type of "stickiness" whereby individuals are resistant to changing their behavior due to the population's inertia. In this model, we observe a nonmonotonicity in the attack rate as a function of various biological and social parameters such as the transmission rate, efficacy of the NPI, costs of the NPI, weight of social consequences of shirking the social norm, and the degree of heterogeneity in the population. We also observe that the attack rate can be highly sensitive to these parameters due to abrupt shifts in the collective behavior of the population. These results highlight the complex interplay between the dynamics of epidemics and norm-driven collective behaviors.

Routine ICU admission after brain tumor surgery: retrospective validation and critical appraisal of two prediction scores

BackgroundRoutine admission to an intensive care unit (ICU) following brain tumor surgery has been a common practice for many years. Although this practice has been challenged by many authors, it has still not changed widely, mainly due to the lack of reliable data for preoperative risk assessment. Motivated by this dilemma, risk prediction scores for postoperative complications following brain tumor surgery have been developed recently. In order to improve the ICU admission policy at our institution, we assessed the applicability, performance, and safety of the two most appropriate risk prediction scores.MethodsOne thousand consecutive adult patients undergoing elective brain tumor resection within 19 months were included. Patients with craniotomy for other causes, i.e., cerebral aneurysms and microvascular decompression, were excluded. The decision for postoperative ICU-surveillance was made by joint judgment of the operating surgeon and the anesthesiologist. All data and features relevant to the scores were extracted from clinical records and subsequent ICU or neurosurgical floor documentation was inspected for any postoperative adverse events requiring ICU admission. The CranioScore derived by Cinotti et al. (Anesthesiology 129(6):1111–20, 5) and the risk assessment score of Munari et al. (Acta Neurochir (Wien) 164(3):635–641, 15) were calculated and prognostic performance was evaluated by ROC analysis.ResultsIn our cohort, both scores showed only a weak prognostic performance: the CranioScore reached a ROC-AUC of 0.65, while Munari et al.’s score achieved a ROC-AUC of 0.67. When applying the recommended decision thresholds for ICU admission, 64% resp. 68% of patients would be classified as in need of ICU surveillance, and the negative predictive value (NPV) would be 91% for both scores. Lowering the thresholds in order to increase patient safety, i.e., 95% NPV, would lead to ICU admission rates of over 85%.ConclusionPerformance of both scores was limited in our cohort. In practice, neither would achieve a significant reduction in ICU admission rates, whereas the number of patients suffering complications at the neurosurgical ward would increase. In future, better risk assessment measures are needed.

An attribute fuzzy concept-oriented three-way utility decision model in multi-attribute environments

In recent years, an increasing number of researchers have focused on how to leverage limited information to make more reasonable decisions, thereby reducing the decision-making risk for decision-makers. Three-way decision, as a method for risky decision-making, offers a new approach to address this challenge and has gained significant attention. In view of this, based on the utility theory, this paper proposes a novel three-way utility decision model oriented to attribute fuzzy concept in multi-attribute environments. Firstly, an attribute fuzzy concept is constructed based on the decision-maker’s minimum requirements on each attribute. Subsequently, a new method for calculating the relative utility function is proposed, which is used to calculate the corresponding decision thresholds. Additionally, a generalized conditional probability calculation method is presented based on the attribute fuzzy concept and the basic semantics of conditional probability. Finally, a new three-way decision model is established to obtain the classification and ranking results of all objects. Through application analysis, comparative study and experimental analysis, the proposed method’s rationality and effectiveness are verified.

Comparative performance analysis of three machine learning algorithms applied to sensor data registered by a leg-attached accelerometer to predict metritis events in dairy cattle

Routinely collected sensor data could be used in metritis predictive modeling but a better understanding of its potential is needed. Our objectives were 1) to compare the performance of k-nearest neighbors (k-NN), random forest (RF), and support vector machine (SVM) classifiers on the detection of behavioral patterns associated with metritis events measured by a leg-attached accelerometer (TrackaCow, ENGS, Hampshire, UK); 2) to study the impact of farm scheduled activities on model performance; and 3) to identify which behaviors yield the highest F1 score for metritis prediction as a function of the number of time window and time-lags. A total of 239 metritis events (188 non-metritis and 51 metritis events) were retrospectively created based on changes in two consecutive uterine evaluations from a dataset containing sensor and clinical data during the first 21 days postpartum between June 2014 and May 2017. These events were associated with a total of 10,874 - 14,138 data points corresponding to hourly measurements of lying time, lying bouts, steps, intake, and intake visits. Sensor data corresponding to the 3 days before each metritis event were aggregated every 24-, 12-, 6-, and 3-hour time windows. Multiple time-lags were also used to determine the optimal number of past observations needed for optimal classification. Similarly, different decision thresholds were compared. Depending on the classifier, algorithm hyperparameters were optimized using grid search (RF, k-NN, SVM) and random search (RF). All behaviors changed throughout the study period and showed distinct daily patterns. From the three algorithms, RF had the highest F1 score, with no impact of scheduled farm activities on classifier performance. Furthermore, 3- and 6-hour time windows had the best balance between F1 scores and number of time-lags. We concluded that steps and lying time can be used to predict metritis using data from 2 to 3 days before a metritis event. Findings from this study will be used to develop more complex prediction models that could identify cows at higher risk of experiencing metritis.

A modified three-event energy detection scheme using decision threshold optimization for sensing performance improvement in a cognitive radio system

A modified three-event energy detection scheme using decision threshold optimization for sensing performance improvement in a cognitive radio system

Parameter Coverage for Testing of Autonomous Driving Systems under Uncertainty

Autonomous Driving Systems (ADSs) are promising, but must show they are secure and trustworthy before adoption. Simulation-based testing is a widely adopted approach, where the ADS is run in a simulated environment over specific scenarios. Coverage criteria specify what needs to be covered to consider the ADS sufficiently tested. However, existing criteria do not guarantee to exercise the different decisions that the ADS can make, which is essential to assess its correctness. ADSs usually compute their decisions using parameterised rule-based systems and cost functions, such as cost components or decision thresholds. In this article, we argue that the parameters characterise the decision process, as their values affect the ADS’s final decisions. Therefore, we propose parameter coverage, a criterion requiring to cover the ADS’s parameters. A scenario covers a parameter if changing its value leads to different simulation results, meaning it is relevant for the driving decisions made in the scenario. Since ADS simulators are slightly uncertain, we employ statistical methods to assess multiple simulation runs for execution difference and coverage. Experiments using the Autonomoose ADS show that the criterion discriminates between different scenarios and that the cost of computing coverage can be managed with suitable heuristics.

P055 Giant Cell Arteritis Diagnosis: Metadiagnosis by robust Bayesian estimate of strength of belief

Abstract Background/Aims Metadiagnosis expresses diagnostic belief as: (i) a numeric probability, (ii) degree of confidence in the probability estimate and (iii) defines the derived information as decision thresholds. We use this approach to stratify suspected GCA patients, by disease probability, into three classes. (a) Likely to have GCA, (b) Unlikely to have GCA, and (c) Uncertain GCA status. Methods We developed a Bayesian algorithm that computes a patient’s disease probability based on their signs, symptoms and reference-standard tests. This algorithm forms the computational engine of the first iteration of an online APP. The computational engine incorporated statistical weights (likelihood ratios) from a meta-analysis of 68 unique studies (14 037 unique patients with suspected GCA). The algorithm requires the specification of a disease prevalence. This is the “prior” probability any patient has GCA. We then update this probability incrementally with each additional item of information elicited from the questionnaire proforma (e.g. female? headache? visual abnormality? …) by incremental evidence aggregation. Finally, the outcome of the patient's reference standard test is incorporated (e.g. positive, negative, indeterminate) and a final posterior GCA probability is computed for each patient. Results Quantitative GCA risk estimate for each individual patient, and how that was altered by the result of reference standard test, produces three distinct groups: a) Low probability group - estimated risk of GCA is significantly lower than the prevalence in the cohort. The reference standard test in this group was uniformly negative.b) intermediate probability group - estimated risk of GCA is around the same as the prevalence in the cohort. The reference standard test (TAB) is almost always negative in this group, but not exclusively.c) high probability group - estimated risk of GCA far exceeds the base or population risk level, and the reference standard test (TAB) is typically confirmatory in this group. Conclusion Applying Bayesian algorithmic metadiagnosis, based on established values of how indicators of disease impact on diagnosis, allows numerical disease prediction and non-mathematically communicating the precision of diagnosis may enhance informed patient choice leading to personalised, targeted intervention. An inbuilt artificial neural network (a deep learning method feedback-loop) trains the algorithm to identify the value of all relevant indicators and so improves predictive accuracy. We plan further studies to establish how this approach could impact on patient well-being and resource utilisation. Disclosure L. Clearkin: None.