An integrated epidemiologic and radiographic algorithm for canine urocystolith mineral type prediction.

Research involved 2 databases. One database (occurrence frequency) comprised the age, breed, gender and urocystolith mineral type (pure chemical types only) from 2041 canine patients submitted to the Minnesota Urolith Center. The other database (imaging) comprised the maximum size, surface (rough, smooth, and smooth with blunt tips), shape (faceted, irregular, jackstone, ovoid, and round) and internal architecture (lucent center, random-nonuniform, and uniform) from 434 canine patients imaged in a urinary bladder phantom. The imaging database was a partial subset of the occurrence frequency database. Imaging techniques simulated were survey radiography and double contrast cystography. The databases were compared using multivariate analysis techniques. Equations were developed to use clinically-relevant characteristics (age, breed, gender, maximum size, surface, shape, and internal architecture) to predict urocystolith mineral types. The goal was to assess the accuracy of the various techniques in predicting the urocystolith mineral types. The combination of signalment (age, breed, gender) and simulated survey radiographic findings does not improve mineral type prediction accuracy (average across all mineral types is 69.9%) beyond that achievable with signalment alone (average across all mineral types is 69.8%). However, the combination of signalment and double contrast cystography does improve mineral type prediction accuracy (average across all mineral types is 75.3%). For comparison, mineral type prediction accuracy without signalment from survey radiographs only was 65.7% across all mineral types. The clinical utility of the algorithm is the option to distinguish urocystolith mineral types requiring surgical vs. medical treatment.