Las dificultades del Derecho Internacional Humanitario como

constraints must be satisfied in order for the transitive trust derivation to be meaning- ful [97]. Opinions can be uniquely mapped to Beta PDFs, and in this sense the fusion operator is equivalent to Bayesian updating. This model is thus both belief-based and Bayesian. Algebraically, a trust relationship between A and B is denoted [A, B], transi- tivity of two arcs is indicated using a binary “:” operator, and the fusion of two parallel paths is indicated with a “” operator. The trust network of Figure 3.3 can then be expressed as:

[A, D] = ([A, B] : [B, D])  ([A, C] : [C, D]) .

The corresponding transitivity operator for opinions is denoted as “⊗” and the cor- responding fusion operator as “⊕”. The mathematical expression for combining the opinions about the trust relationships of Figure 3.3 is then:

ωA D = (ω A B⊗ω B D) ⊕ (ω A C⊗ω C D) .

3.5

Medical Data Trustworthiness Network Structure

Figure 3.4 shows our proposed network structure for deriving the EHR system’s level of trust in received data fields, via our Medical Data Trustworthiness Assessment model. In this section, we explain the functionality of each component. The proto- col by which these components interact is described in Section 3.6.

3.5.1

Healthcare Authority

A Healthcare Authority is a legal body that records information gathered from public sources including, but not limited to, reports received from healthcare providers and medical practitioners about incorrect medical data or procedures, and medical mis- conduct, non-safety, or malpractice cases. The subject of this information is either a healthcare provider, a medical practitioner, or both. The HA uses this information to produce a ratings vector, in which it ranks each reported case according to its severity.

Information from legal public sources about healthcare agents

Healthcare Authority 2 3 D E F Healthcare 1 MDTA _{D E F} Provider MDTA 1 4 5 Reputation Centre

Ratings from the community members

Figure 3.4: Medical Data Reliability Network Structure

This process can be done by applying previously-defined classification rules to each case.

In addition the HA assigns a base rate for each severity rating level and for the prior behaviour of the healthcare agent (either a healthcare provider or medical practitioner). The HA’s ratings vector will have k levels representing the severity (danger) levels for reported cases. Here we assume that level 1 denotes the highest level of severity and level k − 1 the lowest. Level k denotes the special default behaviour for the agent’s community (in the absence of any other information). In this situation we assume ‘perfect’ behaviour of the community (however the receipt of bad ratings may be used by the HA to lower this value).

From this, the HA provides authorised or registered healthcare providers with its opinion (arrow 2 in Figure 3.4) about the medical conduct and practice of a certain healthcare provider or medical practitioner. Here, the HA acts as a Dirichlet repu-

3.5. Medical Data Trustworthiness Network Structure 69

tation system and expresses its trust using the Subjective Logic trust metric opinion.

For example, ωHA_D = (~bHA_D , uHA_D , ~aHA_C ) is the Healthcare Authority’s trust opinion about

healthcare agent D. Vector ~aHA

C represents the HA’s base rate for D’s community (C).

However, the HA does not send its multinomial opinion about an agent to health- care providers in order to preserve the agent’s privacy. Revealing this information will allow the healthcare providers to infer the number of the agent’s bad reported cases. Instead the Healthcare Authority converts its multinomial opinion to a single value by using the point estimate representation in Definition 3.5. Due to the fact that the rat-

ing levels are not evenly distributed, the HA should define a point value vector ~m to

express its weight for each rating level. In this case, the equation in Definition 3.5 is changed accordingly to be:

HA X = k X i=1 ~m(Li) ~S(Li) . (3.1)

3.5.2

Reputation Centre

In our model a Reputation Centre receives ratings from community members (e.g. pa- tients) about healthcare providers and medical practitioners. These ratings are used by the reputation centre to derive a reputation score for those rated agents; this reputation score represents the RC’s subjective opinion (arrow 4 in Figure 3.4). These opinions are communicated to healthcare providers as needed. The RC acts as a Dirichlet repu- tation centre and expresses its opinions in the same way that the Healthcare Authority

does. For example, ωRC

D = (~b RC D , u RC D , ~a RC

C ) is Reputation Centre RC’s opinion about

healthcare agent D.

3.5.3

Medical Data Trustworthiness Assessment Service

The Medical Data Trustworthiness Assessment service is employed by the Electronic Health Record system to measure the reliability of medical data sourced from other healthcare agents (e.g. healthcare providers or medical practitioners). The EHR sys- tem has a database that records its experiences with other healthcare agents. These

experiences are created from the reports that are received from the EHR system’s users about received external medical data. The EHR system uses this information to either record positive or negative experiences with the agents who created these data. The EHR system’s experiences are then used by the MDTA service, that acts as a Beta reputation system, to compute the EHR system’s opinion about a certain healthcare agent. The EHR system’s opinion about a healthcare agent D (arrow 1 in Figure 3.4)

is denoted as ωEHR? D = (b EHR? D , d EHR? D , u EHR? D , a EHR? C ).

In addition, the MDTA service can communicate with the HA and the RC to get their opinions about a certain agent, for use in the MDTA srevice’s reliability calcula- tion process. Also, the MDTA service maintains dynamic opinions about the HA and the RC (arrows 3 and 5 in Figure 3.4) that are calculated based on opinion compari- son [94].