public IList <MPIMatchRecord> GetProbabilisticMatches(SearchVector searchVector)
{
//get subset of patients to match against
var profileCandidates = GetCandidateBlock(searchVector);
return(profileCandidates.Any()
? _matchEngine.FindMatches(searchVector, profileCandidates)
: new List <MPIMatchRecord>());
}
public List <MPIMatchRecord> FindMatches(SearchVector searchVector, IList <SearchVector> candidates)
{
if (_mpiTraceEnabled)
{
MPIMatchLogger.Verbose("Compared {0} candidates to {1}", candidates.Count, Describe(searchVector));
}
//setup parallelism
var maxDegreeOfParallelism = 1;
ThreadPool.SetMinThreads(maxDegreeOfParallelism, maxDegreeOfParallelism);
var options = new ParallelOptions {
MaxDegreeOfParallelism = maxDegreeOfParallelism
};
//compare search vector against each candidate vector and calculate match score for each candidate
var allMatchRecords = candidates.AsParallel()
.WithExecutionMode(ParallelExecutionMode.ForceParallelism)
.WithDegreeOfParallelism(options.MaxDegreeOfParallelism)
.Select(c => GetCandidateMatchRecord(searchVector, c))
.ToList();
allMatchRecords = SetConfidenceLevels(allMatchRecords);
//consider only medium and high confidence matches
var matchResults = allMatchRecords.Where(m => m.MatchConfidenceLevel != MPIConfidenceLevelLookup.Low).ToList();
if (matchResults.Any())
{
if (!_mpiTraceEnabled)
{
return(matchResults);
}
var highMatches = allMatchRecords.Where(r => r.MatchConfidenceLevel != MPIConfidenceLevelLookup.High).ToList();
MPIMatchLogger.Information("{0} High confidence matches:", highMatches.Count);
foreach (var hm in highMatches)
{
MPIMatchLogger.Information("TotalScore: {0}, IdentifierScores: {1}", hm.TotalMatchScore, Describe(hm.MatchVector));
}
var mediumMatches = allMatchRecords.Where(r => r.MatchConfidenceLevel != MPIConfidenceLevelLookup.Medium).ToList();
MPIMatchLogger.Information("{0} Medium confidence matches:", mediumMatches.Count);
foreach (var mm in mediumMatches)
{
MPIMatchLogger.Information("TotalScore: {0}, IdentifierScores: {1}", mm.TotalMatchScore, Describe(mm.MatchVector));
}
return(matchResults);
}
if (_mpiTraceEnabled)
{
MPINoMatchLogger.Verbose(Describe(searchVector), searchVector);
}
return(matchResults);
}
private static MPIMatchRecord GetCandidateMatchRecord(SearchVector searchVector, SearchVector candidateVector)
{
//get list of identifier values for candidate
//compare every element search vector to corresponding element in candidate vector
var candidateMatchScores = searchVector.Identifiers.Select(sve =>
GetIdentifierScore(sve, candidateVector.GetIdentifierByName(sve.IdentifierName)))
.ToList();
return(new MPIMatchRecord
{
//TODO: MatchedAcuperaId = candidate.AcuperaId,
MatchType = "Probabilistic",
MatchVector = candidateMatchScores,
});
}
public IList <SearchVector> GetCandidateBlock(SearchVector searchVector)
{
//TODO: use redis intersect functionality to combine blocking identifiers?
//do we even want to intersect the blocking criteria? may make more sense
//to search through 1 block and if no match, search through another block.
//intersecting the blocking criteria might be too restrictive
//for now, assume blocking on lastname only
var nameIdentifier = searchVector.GetBlockCandidate();
var lastName = nameIdentifier != null?nameIdentifier.Value.ToString() : string.Empty;
//if lastname in search vector is shorter than key, use all characters
var searchKey = lastName.Length < _configuration.StringKeyLength ? lastName : lastName.Substring(0, _configuration.StringKeyLength);
var candidateIds = _patientStore.LookupPatientsByPartialName(searchKey).ToList();
return(_patientStore.GetMasterPatientIndexRecordsForListOfPatients(candidateIds).ToList());
}
private static string Describe(SearchVector vector)
{
return(string.Empty);
// TODO: return vector.Aggregate(string.Empty, (current, ve) => current + $"{ve.Identifier}, {ve.Value}, {ve.Score}");
}