Limitations from clinical data
There are a few limitations noticed during the construction of the simulation model. First of all, studied populations in clinical trial data was not Norwegian; however, distribution of Norwegian T2DM patients was applied when defining the simulation population. Secondly, the longest duration of head-to-head clinical trial data available was for three years, comparing bariatric surgery to medical therapy with or without lifestyle changing programs. A conservative approach has been adopted for the changes of clinical parameters as well as disease progression after three years. Assumption is made that the corresponding parameters do not fluctuate after three years.
Furthermore, the probability of having diabetes relapse is unknown due to the limited follow-up duration. Hence, the resultant ICERs could be over-estimated even though the best available data was applied.
Limitation from UKPDS Outcomes Model and its risk equations
Another limitation comes from the use of risk equations extracted from UKPDS Outcomes Model. The model was developed and published in 2004 (P. M. Clarke et al., 2004), based on the data collected from UK T2DM patients between 1977 and 1997 (Hayes et al., 2013) and version 2 was re-developed in 2013, with the addition of new clinical data from its 10-year post-trial monitoring period (Holman, Paul, Bethel, Matthews, & Neil, 2008). Although the data could be outdated, UKPDS study is still considered as the largest and longest study on type 2 diabetes mellitus patients (Genuth et al., 2003).
The model is mediated by eight risk equations and four mortality equations, and thus it can estimate
63 the likely occurrence of a range of complications over patients’ lifetimes (P. M. Clarke et al., 2004).
However, these predictive equations were estimated using data from UK T2DM patients, and therefore, it could be a limitation of this study as Norwegian risk data is preferable.
Besides, the UKPDS Outcomes Model simulates macro- and microvascular complications only, i.e. CHF, IHD, MI, stroke, ulcer, amputation, renal failure and blindness, where other minor complications and pharmacologically induced hypoglycaemia were not included. As there is no relevant data of likelihood occurrence of these complications after bariatric surgery, they are excluded in this study. It is expected that the frequency pharmacologically induced hypoglycaemia and associated costs and disutility would be reduced while the dosage of oral antidiabetic drugs or insulin was cut after surgery. In such way, ICER of bariatric surgery over lifestyle intervention could be underestimated despite the low cost of hypoglycaemia event.(Alsumali et al., 2018)
Limitations from cost data
Various sources of cost data were applied in the model, and this may subject to inconsistency and transferability issues. During the selection process, data from studies conducted in Norway and other Scandinavian countries was preferred due to the similarity of healthcare system among them (J.
Magnussen, Vrangbaek, & Saltman, 2009). Therefore, the cost data for bariatric surgery, lifestyle intervention and most of the T2DM complications were taken from reports issued by Norwegian Institute of Public Health (Liv Giske, 2018; Pike et al., 2013; Wisløff et al., 2008). Furthermore, published Norwegian cost data for T2DM at the time of analysis were scant. Two Norwegian studies have examined the cost of illness for T2DM; but both studies have pooled both type 1 and type 2 diabetes mellitus patients and the result may not be representative of the T2DM population in Norway (Solli, Jenssen, et al., 2010; Sorensen et al., 2016). Therefore, diabetes medication costs were obtained from a Swedish study (Henriksson et al., 2000). On the other hand, these studies did not report costs by severity of the disease even it is known that treatment of T2DM would become more costly with disease progression as more insulin would be prescribed with progression (Solli, Jenssen, et al., 2010).
64 Moreover, post-operative clinical outcomes could vary widely across hospitals and surgeons (Birkmeyer et al., 2013). One study has reported long-term medical complication associated with bariatric surgery in Norway (Jakobsen et al., 2018), but unfortunately the time of occurrence was not mentioned. Hence, probability of post-operation complications and associated costs were obtained from a recent CEA of bariatric surgery in the setting of US (Alsumali et al., 2018).
Apart from that, the dosage of medication, including anti-diabetic medication and lipid-lowering drug, was assumed to be halved when HbA1c level dropped to below 7% for consecutive two years owing to simplification. However, dosages of specific medication shall be correlated to measures of biological parameters and adjusted regularly in the reality. Therefore, the medication costs after bariatric surgery could be over-estimated due to the strict assumption. Moreover, it was assumed that there is none side effect come with lifestyle intervention.
Limitations from utility data
Utility data can be divided into four parts, i.e. disutility along with the interventions, disutility caused by T2DM complications, utility of T2DM patients and utility after diabetes remission. Concerning disutility along with bariatric surgery, it was extracted from the same CEA done in US (Alsumali et al., 2018). However, there is no available data for disutility associated with surgical complications. Thus, the QALYs gained for patients undergone surgery may be over-estimated. At the same time, it was assumed that no disutility caused by the comparator, lifestyle intervention.
Apart from that, disutility due to T2DM complications were mainly based on two European studies (Bagust & Beale, 2005; P. Clarke et al., 2002). An early review paper on health utilities (Beaudet et al., 2014) concluded that the aforementioned articles were the preferred utility value set in accordance with the UK National Institute for Health and Care Excellence (NICE) because of the large sample size and use of EQ-5D questionnaire. Also, taking as many utility values as possible from one or two studies helps to minimize inconsistency internally.
65 For utility value of T2DM patients, it was replaced with the Norwegian one reported earlier by Solli (0.850) et al (Solli, Stavem, et al., 2010) as it is much higher than utility value (0.785) proposed by Clarke et al (P. Clarke et al., 2002). Nevertheless, there is no severity specific utility for T2DM patients.
Due to the unavailability of T2DM remission utility, utility of Danish population norm was applied (Janssen & Szende, 2014).
Implications
The discrete event simulation model has simulated the natural history and outcomes at individual patient level. It is indicated for the Norwegian type 2 diabetes mellitus patient population with BMI within 25 – 35 kg/m2. It has demonstrated that bariatric surgery versus lifestyle intervention result in better health outcomes with higher costs; and bariatric surgery is considered as cost-effective under Norwegian WTP threshold. This cost-utility analysis may influence the decision makers to extend the surgical option to T2DM patients without morbidly obesity.