Working Papers 2013

5/2013 José C. Ferrão, Mónica D. Oliveira, Filipe Janela, Henrique M. G. Martins, "Clinical coding based on structured HER systems - a supervised learning approach using routinely collected data”

Objectives: clinical coding is an essential process whereby health data is indexed using standard terminologies for billing and reporting. To mitigate workload and errors, partial automation has been seek using free-text data from electronic health records (EHR), but its use is difficult to generalize in contexts with different scopes and languages. In this work, we propose an approach to support clinical coding using fully structured EHR data.  

Methods: we propose a methodology encompassing EHR data processing to define a feature set and a supervised learning approach to predict ICD-9-CM code assignment. We employ a fast correlation-based filter to reduce dimensionality and binary relevance method to transform the multi-label problem into multiple binary classification problems. Four supervised learning models – decision trees, naïve Bayes, logistic regression and support vector machines – were tested and compared.

Results: tests performed with a real dataset yielded F1 scores of 0.48, 0.58, 0.54 and 0.53 for decision trees, naïve Bayes, logistic regression and support vector machines, respectively. Performance varies greatly across codes and appears to be related to the clinical concepts underlying each code rather than to the supervised learning method employed.  

Conclusions: the use of structured EHR data to support clinical coding shows promising results. The analyses carried out in this work indicate lines for further improvement, namely data quality assessment and improvement and the inclusion of expert knowledge in model revision and validation.

4/2013 Teresa Cardoso, Mónica D. Oliveira, Ana Barbosa-Póvoa, Stefan Nickel, "Planning a long-term care network with uncertainty, strategic policy and equity considerations: A stochastic planning approach"

Departing from a structuring of key uncertainties and of policy options inherent to the reorganization of a long-term care (LTC) network, this study proposes a stochastic mixed integer linear programming (MILP) model for planning the delivery of such a network. The model assists health care planners on how to plan the delivery of the entire range of LTC services – from institutional to home-based and ambulatory services – when their main policy objective is the minimization of expected costs and they consider that satisficing levels of multiple dimensions of equity need to be respected. Equity dimensions (modeled as constraints) include equity of access, equity of utilization, socioeconomic equity and geographical equity. The proposed model provides planners with key information on when and where to locate services and with which capacity, how to distribute this capacity across services and patient groups, and which changes to the network of care are needed over time (increasing and reducing capacity, and the opening and closure of services). Model outputs take into account the uncertainty surrounding LTC demand, and vary according to strategic health policy options adopted by governments. The applicability of the model is demonstrated through the resolution of a case study in the Great Lisbon region in Portugal. Results illustrate how the LTC network should evolve when different strategic health policy options are adopted, and provide estimates of the expected costs of achieving satisficing equity levels.

3/2013 Teresa C. Rodrigues, Carlos A. Bana e Costa, Mónica D. Oliveira , “Multicriteria Cognitive Map: A tool for problem-structuring and multicriteria evaluation”

In complex decision making situations, the need for tools to support both problem structuring and multi-criteria evaluation of decision alternatives arises. This study proposes a new tool – Multicriteria Cognitive Map (MCM) – to help constructing a multicriteria evaluation model. A MCM combines, in a coherent and integrated way, Cognitive Mapping and concepts and methods from Multicriteria Decision Analysis, supporting both problem structuring and the evaluation of decision alternatives. Developing a MCM integrates two recursive phases: a first phase to structure the problem by capturing the issues and their systemic relationships in a means-ends network, whereby each node represents a concept/idea and each means-ends link between two nodes represents influence; and a second phase to determine the impact that alternatives (means) have in the values (ends) of the decision makers. This latter phase uses the MACBETH (Measuring Attractiveness by a Categorical Based Evaluation Technique) approach to measure the strength of the influence of each means-end link. In order to illustrate the applicability and usefulness of the MCM tool to evaluate decision alternatives, a small but intuitive example of an evaluation model is described, as well as it is reported a real application of MCMs in the process of constructing a health index based on a multicriteria evaluation model structure.


2/2013 Diana F. Lopes, Carlos A. Bana e Costa, Mónica D. Oliveira, Alec Morton, “Using MACBETH with the Choquet Integral Fundamentals to Model Interdependencies between Elementary Concerns in the Context of Risk Management”

Effective risk management typically requires the evaluation of multiple consequences of different sources of risk, and multicriteria value models have been used for that purpose. The value of mitigating a risk impact is often considered by risk managers as dependent on the levels of other impacts, therefore there is a need for procedures to identify and model these interactions within a value measurement framework. The Choquet Integral (CI) has been used for this purpose, and several studies in the performance measurement literature have combined the 2-additive CI operator with the MACBETH approach to model interdependencies in real contexts. In this paper, we propose an alternative procedure to model interdependencies and determine the CI parameters from one single MACBETH global matrix. The procedure is illustrated with the construction of a descriptor of impacts to evaluate the risk impacts at ALSTOM Power. The paper further explains the questioning protocol to apply the proposed procedure, as well as how decision-makers can interpret the CI parameters.

1/2013 Mónica D. Oliveira, Carlos A.  Bana e Costa, Diana F. Lopes, "Improving risk matrices using multicriteria and portfolio decision analysis"

Risk matrices are adopted and recommended by many organizations, but risk analysis literature has shown that the way they are usually constructed violates some basic theoretical principles, giving rise to inconsistent risk ratings. This article studies ways in which multiple criteria and portfolio decision analyses can improve the design and the deployment of risk matrices. Firstly, it introduces ‘value risk-matrices’, built with the MACBETH multicriteria method in three modeling steps: (1) building a value measurement scale on each impact dimension and constructing of a subjective probability scale, (2) additive aggregation of the value scales into a cumulative value scale, and (3) design of the value risk-matrix based on the cumulative value scale, the subjective probabilities scale and the identification of risk categories. The rating of risks by risk managers, using risk matrices, is the basis for the identification of risk mitigation actions. The prioritization of actions in face of limited resources is addressed in the last part of the article, making use of the recent portfolio decision analysis component of the MACBETH decision support system. Taken all together, the article sketches a new modeling approach for Improving Risk Matrices (IRIS), which is illustrated with data from a case study developed at ALSTOM Power.