National Institute for Quality- and Organizational Development in Healthcare and Medicines COMPARING THREE DIFFERENT METHODS OF HALF-CYCLE CORRECTION Bertalan Németh, Vera Szekér National Institute of Quality- and Organizational Development in Healthcare and Medicines, Department of Health Technology Assessment, Budapest, Hungary Introduction The Department of Health Technology Assessment of the National Institute for Quality- and Organizational Development in Healthcare and Medicines (GYEMSZI TEI) is responsible for the critical evaluation of the reimbursement submissions of medicines and medical devices in Hungary. The calculations in these submissions are often based on Markov models and some of these models use half-cycle correction. Half-cycle correction is a method used in health economics in Markov models to improve the accuracy of the results. The basic idea behind this method is that the flow of patients between Markov states is continuous therefore if the number of patients is measured at the beginning or at the end of each cycle the result can be overestimated or underestimated. By using half-cycle correction the difference from real data can be reduced.1 See Figures 1 to 4. Figure 7 The flow of patients in case A of our model. Figure 9 The flow of patients in case C of our model. Figure 1 The flow of patients in a state of a Markov model. Figure 2 Determining the number of patients at the end of each cycle. Figure 8 The flow of patients in case B of our model. Figure 10 The flow of patients in case D of our model. Results Table 1 shows the detailed results of our model in case A. It can be seen that without using one of the three half-cycle correction methods the error of the results is generally higher. Of the three half-cycle correction methods Simpson's rule was the most accurate but the difference from real data was less than 1% in almost every case when we used any half-cycle correction method. Please note that our model is based on a hypothetical set of paramaters so the exact numbers written in table 1 are less important than their order of magnitude which we highlited with four different colors. Table 1 The difference from the actual data in case A of our model. Figure 3 Determining the number of patients at the beginning of each cycle. Figure 4 Determining the number of patients by sing half-cycle correction. Last year we examined the theoretical concept of half-cycle correction.2 Our conclusion was that it is advised to use half-cycle correction when the number of cycles is small and the length of cycles is too long and there is no possible way to reduce it. Our conclusion was in line with the findings of Barendregt3 and Naimark et al.4 It is worth to note that in some rare cases half-cycle correction can increase the difference from real data in the model. See Figures 5 and 6. In this poster we aim to assess the relative performance of the alternatives to the standard half cycle correction and to account for the connection between health states in a Markov model by using a model we developed. All of our findings are based on our hypothetical Markov states. Figure 5 Determining the number of patients at the end of each cycle. Figure 6 Determining the number of patients by using half-cycle correction. Note the higher difference from real data compared to Figure 5. Decription of the model Our model is a 5-state Markov model where the number of patients in health states follow different courses. Each health state represented a typical shape for the flow of patients. The first two health states were the ones where the number of patients was monotonously decreasing or increasing. In the third health state the number of patients was increasing at the beginning but began decreasing after a certain point. We added a health state with a continuously oscillating number of patients and one where the number of patients remained the same after the first cycle. We chose to compare three different half-cycle correction methods. We used the basic half-cycle correction method, Simpson's method, and the mid-cycle method. It is important to note that the description of the simple half-cycle correction method can slightly differ in the sources. In most cases the two following descriptions are used 5: In Figure 11 we compared the results from the five different health states in four different cases. It can be seen that in cases A, C and D determining the number of patients at the beginning or at the end of each cycle resulted in a higher difference from real data than using the three half-cycle correction methods which didn't differ from each other significantly. The only exception is case B when the standard half-cycle correction method resulted in a higher difference from real data than determining the number of patients at the end of each cycle but the other two half-cycle correction methods still resulted in more accurate results. This case was similar to the hypothetical case we presented in Figures 5 and 6 where the number of patients in each health state shifted significantly during the first cycle and changed less during later cycles. It can also be mentioned that the difference from real data was higher in every case and with every method if the number of cycles was lower. Membership is counted at the beginning of each cycle Membership is counted at the end of each cycle Standard half-cycle correction Simpson’s method Mid-cycle method Figure 11 The difference from the actual data after 40, 30, 20 and 10 cycles in the four cases of our model. In table 2 we showed the method which was the most accurate in every case in every health state. It can be seen that in 78 of the 80 possible cases Simpson's method and the mid-cycle method was the most accurate and the standard half-cycle correction method resulted in the smallest difference from real data in the remaining two cases. It can also be mentioned that Simpson's rule resulted in the smallest difference from real data in the health states where the flow of patients was decreasing, hill shaped or wave shaped. Table 2 The half-cycle correction method with the smallest difference from the actual data. - Add half of the number of patients at the end of the first cycle to the total patient population but subtract half of the number of patients at the last cycle. This method means that we establish two cycles with half of the original cycle length and between them there will be cycles with normal cycle length. Green: standard half cycle correction method. - Take the average of the patient numbers at the beginning and at the end of each cycle. This is also called the life table method. Purple: Mid-cycle method. Blue: Simpson's rule. It can be shown that the two methods equal on the total number of patients but with using the life table method the dynamics of patient flow is more similar to the real data.3 The mid-cycle method is when we take into account the number of patients at the middle of each cycle. Simpson's rule is a method used for numerical integration. In our model we used the definition when we add the number of patients at the beginning and at the end of a cycle and add four times the number of patients at the middle of each cycle then divide the sum by six. For comparison we included the two cases when the number of patients is measured at the end or at the beginning of each cycle. We examined the results after 10, 20, 30 and 40 cycles and we ran the model with four different sets of parameters. In these four cases the flow of patients in each health state differs significantly. See Figures 7 to 10. We determined the exact number of patients in each health state by using numerical integration and compared the results of each method to these numbers. Conclusion Based on the results of our model the most accurate method for half-cycle correction is Simpson's method as in most cases it differed the least from the actual data. It's important to note that in most cases the difference from real data was similar when we applied the other two methods, the standard half-cycle correction method or the mid-cycle method. With one exception the accuracy of the model was signficantly increased by using any half-cyle correction method in every case we examined compared to cases when no half-cycle correction method was applied. The difference from real data was also higher in every case and with every method when the number of cycles was lower. Literature 1. Naimark DM, Bott M, Krahn M. The half-cycle correction explained: two alternative pedagogical approaches. Medical Decision Making. 2008 Sep-Oct; 28(5):706-12. 2. Nemeth B, Vincziczki Á. The Role of Half-Cycle Correction in the Models Used for Health Technology Assessment. Poster presentation at the ISPOR 16th Annual European Congress. 3. Barendregt JJ. The half-cycle correction: banish rather than explain It. Medical Decision Making. 2009 jul-Aug; 29(4):500-2. 4. Naimark DM, Kabboul NN, Krahn MD. The half-cycle correction revisited: redemption of a kludge. Med Decis Making. 2014 Apr; 34(3):283-5. 5. Clarke PM, Wolstenholme JL, Wordsworth S. Applied methods of cost-effectiveness analysis in healthcare. Oxford University Press; 2010. p. 224-5.