Assessing accuracy and precision of developed equations in prediction of blood lithium levels in psychiatric patients in Khon Kaen Rajanagarindra Psychiatric Hospital

Abstract

Introduction: Lithium is commonly used as a mood stabilizer in psychiatric patients with mood disorders, especially for the treatment of bipolar disorder. However, lithium has narrow therapeutic index and may cause dangerous toxicity on kidneys. Therefore, optimal dose of lithium is very important to ensure an effective therapy and to prevent its toxicity. In clinical practice, monitoring of blood lithium levels is required. Several methods have been proposed for the dosage determination. One is a priori prediction method that uses patient’s characteristics in mathematic equations for prediction of lithium dose or blood lithium level in the patient. The objective of the present study was to determine the accuracy and precision of 4 developed equations in prediction of blood lithium levels in psychiatric patients in Khon Kaen Rajanagarindra Psychiatric Hospital. Methods: With specific inclusion criteria, data were collected retrospectively from the medical records of patients who received lithium between January 2011 and December 2012. Predicted blood lithium levels were computed from 4 equations: (1) Tearo et al equation, (2) Abou-Auda et al equation, (3) Chiu et al equation, and (4) Huang et al equation. By comparing predicted blood lithium levels from each equation with the actual levels, accuracy of each equation was assessed by computing mean prediction error and percent error and the precision was assessed by computing root mean square error. In addition, correlation coefficients between predicted blood lithium levels and the actual levels were computed for each equation. Sensitivity and specificity tests of each equation were also calculated. Results: Of 620 inpatients who used lithium, only 90 met the inclusion criteria. For accuracy evaluation, the lowest mean prediction error was in Tearo et al equation (median [interquartile range], -0.010 [-0.120 to 0.150] mmol/L; percent error, -1.66% [-21.37% to 27.71%]) compared with Abou-Auda et al equation (-0.685 [-1.0200 to -0.288] mmol/L; -128.23% [-209.54% to -43.24%]), Chiu et al equation (0.110 [-0.040 to 0.275] mmol/L; 21.52% [-6.01% to 68.04%]), and Huang et al equation (-0.175 [-0.300 to -0.045] mmol/L; -33.63% [-47.87% to -13.90%]). For precision evaluation, the lowest root mean square error was in Tearo et al equation (median [interquartile range], 0.141 [0.000 to 0.250]) compared with Abou-Auda et al equation (0.696 [0.312 to 1.100]), Chiu et al equation (0.173 [0.075 to 0.350]), and Huang et al equation (0.173 [0.100 to 0.300]). The greatest Spearman’s rho correlation coefficient between the predicted blood lithium levels and actual blood lithium levels was in Tearo et al equation (=0.666; p-value<0.01) compared with Abou-Auda et al equation (= 0.641; p-value<0.01), Chiu et al equation (= 0.526; p-value<0.01) and Huang et al equation (= 0.438; p-value<0.01). Sensitivity and specificity tests were done to determine the performance of each equation in correctly classifying the patients with their blood lithium levels as being inside the therapeutic range (0.6-1.2 mmol/L). Sensitivity tests of Tearo et al, Abou-Auda et al, Chiu et al, and Huang et al equations were 72%, 8%, 78% and 14%, respectively. Specificity tests of Tearo et al, Abou-Auda et al, Chiu et al, and Huang et al equations were 77%, 90%, 59% and 89%, respectively. Conclusion: Of all 4 equations assessed, Tearo et al equation is the best in accuracy and precision for prediction of blood lithium levels in psychiatric patients in Khon Kaen Rajanagarindra Psychiatric Hospital.