On 2015 Sep 20, Ralph Brinks commented:

Most guidelines in health economic modeling stress the importance of the quality of the input data for the results of an economic model [1]. The input data of the transition probabilities between the BMI groups in the model of this work is based on empirical data from Germany and a method, which is stated to be described in [3]. Essentially, the transition probabilities are estimated from age-specific prevalence data, which is a well-known problem in epidemiology [4]. The method used in [2], however, is inappropriate, which I will demonstrate by showing the magnitude of the error. Terminology here is similar to [2], the BMI groups are numbered: normal (1), overweight (2), obese (3). The transition probability from state x to state y is denoted by tpxy

Given the data of 53 year old males tp12 = 0.02159, tp23 = 0.01035, [2], and assumed that the probability of dying is 0.006 for all BMI groups [5], then at age 53 the prevalences in the respective BMI groups are about P(53) = (0.45, 0.3, 0.25), see Figure 2 of [2]. After simulating one cycle of the Markov model, we obtain the prevalence at age 54 as P(54) = (0.4402, 0.3067, 0.2531). If we now use Equation (1) in [2] to estimate tp12, we obtain (0.3067-0.3)/0.45 = 0.01489. Compared to tp12 = 0.02159, this is an relative error of -31%, a magnitude which exceeds the relative errors in the sensitivity analysis (+/-20%) by far. This is not acceptable.

Thus, I query whether the results of [2] are valid. Instead of using the inappropriate Equations (1) and (2) in [2] I suggest to use studies surveying the transition probabilities between BMI groups directly.

[1] Penaloza Ramos MC, Barton P, Jowett S, Sutton AJ. A systematic review of research guidelines in decision-analytic modeling. Value Health 2015; 18: 512-29.

[2] Sonntag D, Ali S, Lehnert T, Konnopka A, Riedel-Heller S, König HH. Estimating the lifetime cost of childhood obesity in Germany: Results of a Markov Model. Pediatr Obes 2015 [Epub ahead of print]

[3] Miller DK, Homan SM. Determining transition probabilities: confusions and suggestions. Med Dec Making 1994; 14: 52-58.

[4] Keiding N. Age-specific incidence and prevalence: a statistical perspective. J R Statist Soc A 1991; 154: 371-412

[5] Statistisches Bundesamt. Generationensterbetafeln Deutschland: Wiesbaden 2011.

On 2015 Sep 20, Ralph Brinks commented:

Most guidelines in health economic modeling stress the importance of the quality of the input data for the results of an economic model [1]. The input data of the transition probabilities between the BMI groups in the model of this work is based on empirical data from Germany and a method, which is stated to be described in [3]. Essentially, the transition probabilities are estimated from age-specific prevalence data, which is a well-known problem in epidemiology [4]. The method used in [2], however, is inappropriate, which I will demonstrate by showing the magnitude of the error. Terminology here is similar to [2], the BMI groups are numbered: normal (1), overweight (2), obese (3). The transition probability from state x to state y is denoted by tpxy

Given the data of 53 year old males tp12 = 0.02159, tp23 = 0.01035, [2], and assumed that the probability of dying is 0.006 for all BMI groups [5], then at age 53 the prevalences in the respective BMI groups are about P(53) = (0.45, 0.3, 0.25), see Figure 2 of [2]. After simulating one cycle of the Markov model, we obtain the prevalence at age 54 as P(54) = (0.4402, 0.3067, 0.2531). If we now use Equation (1) in [2] to estimate tp12, we obtain (0.3067-0.3)/0.45 = 0.01489. Compared to tp12 = 0.02159, this is an relative error of -31%, a magnitude which exceeds the relative errors in the sensitivity analysis (+/-20%) by far. This is not acceptable.

Thus, I query whether the results of [2] are valid. Instead of using the inappropriate Equations (1) and (2) in [2] I suggest to use studies surveying the transition probabilities between BMI groups directly.

[1] Penaloza Ramos MC, Barton P, Jowett S, Sutton AJ. A systematic review of research guidelines in decision-analytic modeling. Value Health 2015; 18: 512-29.

[2] Sonntag D, Ali S, Lehnert T, Konnopka A, Riedel-Heller S, König HH. Estimating the lifetime cost of childhood obesity in Germany: Results of a Markov Model. Pediatr Obes 2015 [Epub ahead of print]

[3] Miller DK, Homan SM. Determining transition probabilities: confusions and suggestions. Med Dec Making 1994; 14: 52-58.

[4] Keiding N. Age-specific incidence and prevalence: a statistical perspective. J R Statist Soc A 1991; 154: 371-412

[5] Statistisches Bundesamt. Generationensterbetafeln Deutschland: Wiesbaden 2011.