Stem cells and beta cell replacement therapy: a prospective health technology assessment study

The form of beta cell replacement therapy was modeled as a series of identical re-extractable subcutaneous implant devices (‘sheets’). Each of these devices contain hES-derived cells, specifically pancreatic progenitors which were modified to attain the function of beta cells. Those cells are enclosed within a casing which shields them from the immune system but allows the transport of nutrients to and hormones from the encapsulated cells [14, 15]. In our study we make the important assumption that this shielding effect completely removes the need for immunosuppressive medication. Patients in our study can get up to four transplantations [20]. The average proportion of patients with full graft function after each transplantation was assumed to be increasing from first to third and fourth transplantation, i.e. from 15% to 70% and then 85% for the third and fourth ones [20].

The comparator treatment is intensive insulin therapy involves frequent self-monitoring of blood glucose and multiple daily insulin injections, further details are described elsewhere [20].

We conducted a probabilistic and structural sensitivity analysis to investigate the cost-effectiveness of stem cell-based beta cell replacement therapy and to evaluate uncertainty around our results. Our simulation model was a discrete state-transition Markov model, which had a lifetime horizon. Its hypothetical cohort was composed of type 1 diabetes patients with hypoglycemia unawareness in the province of Alberta who fulfill the inclusion criteria to get an islet transplant. The model took the perspective of the provincial healthcare provider and its inputs were the same as in the pre-existing cost-effectiveness model [20], except for the described variations. Effectiveness was expressed in quality-adjusted life-years (QALYs) to measure the impact of therapy on both quality of life and life expectancy. All monetary estimates are expressed in 2016 Canadian dollars, with necessary adjustments made using the Canadian consumer price index for health and personal care [22, 23].

We updated model parameters from our pre-existing model of unstable type 1 diabetes [20] as described below. We model a future technology functioning without the need of immunosuppression. Therefore we had to change the parameters that – even partially - had to do with this medication. For that we removed all disutilities, costs and probabilities that had only to do with immunosuppression. The parameters for rate, costs and disutility of initial complications were adjusted by lowering each by 40% because this portion was attributed solely to immunosuppression. Specifically the rate changed from 0.65 to 0.39, the cost from $600 to $360, and the disutility from 0.05 to 0.03. Further, a study of the impact of type 1 diabetes complications (N = 2341) served us to update our utility i.e. quality of life estimates [24]. Patients in that study were more comparable to our hypothetical cohort than the ones in our original data sources. Yet they were still younger (39.3 vs. 47.0 years old) and with shorter diabetes duration (16.3 vs. 29.4 years) [20, 24]. Given that new evidence, we included neuropathy in the diabetes-related complications and adjusted our overall estimate for the complications state not only for multiple complications, but also to fit the actual age and duration of diabetes in our cohort [2, 24‐29]. The utility parameter in the complications state was therefore adjusted from 0.57 to 0.47.

The cost-effectiveness model was constructed and run with the software TreeAge Pro 2016 (Williamstown, MA, USA). The cost of goods modeling was constructed and run using Microsoft Excel. Costs and benefits were discounted at 3%, and sensitivity analysis were performed at 0% and 5%. These were the rates that had been recommended by the Canadian Agency for Drugs and Technologies in Health [30]. Half-cycle correction was applied. Our probabilistic analyses used 64,000 iterations for each scenario. We estimated the value of further research reducing the decision uncertainty by way of value-of-information analysis [31, 32]. We calculated the expected value of perfect information (EVPI) and the expected value of partial perfect information (EVPPI) for the cost of goods group of parameters. For that EVPPI calculation we used nested Monte Carlo simulations with 600 ‘outer’ and 600 ‘inner’ loops. Additional information on our value of information approach, including choice of WTP thresholds, can be found elsewhere [20].

To integrate the results of the cost of goods modeling into our cost-effectiveness model, we took the parameter representing the cost per transplantation within the transplant state in the original model, and split it up into non-dose costs and dose costs. Based on literature we assumed the non-dose costs, including the transplant procedure, to be about 38% of the costs per transplantation [33]. For that parameter of our model we used a Gamma distribution with a relative standard deviation of 10% (i.e. standard deviation as percentage of the mean). For the dose costs, based on our cost of goods (COG) model we used the following equation:

Here dose costs are calculated multiplying the cost of goods upstream by a cost of goods downstream factor and a factor that we called “regulatory burden factor”. The cost of goods upstream came from the above described fitted distributions estimating the “pure” production cost of the cells. The downstream factor accounted for the so-called downstream processing, which is necessary after the cells are produced, e.g. cell harvesting, volume reduction, washing, formulation for storage or delivery (see Fig. 2). We used the regulatory burden factor to account for the possibility of additional regulatory burden due to stricter regulatory requirements for a new cell production process whose product is designed to enter regular healthcare practice.

The mean of the cost of goods downstream factor (multiplier) was assumed to be three, four and eight, depending on scenario, based on expert opinion and published literature [34]. The mean of the regulatory burden factor was assumed to be 1.2 and 1.8 depending on scenario (i.e. 20% and 80% additional costs respectively due to regulation). Both multipliers were made probabilistic using a Log-normal distribution and expert opinion on variation estimates.

We modeled the cost of goods in Microsoft Excel based on a report describing manufacture of pancreatic progenitors from single cell cultures of hES cells [6]. Briefly, the modeled process consisted of thawing one or more frozen hES cells vials and expanding them in adherent culture for about 14 days, passaging four times. Then, hES cells were cultured in suspension forming cell clusters, while a cocktail of molecular signals was added to the media to promote stepwise differentiation of hES cells into pancreatic progenitors.

The cost of goods was estimated by adapting a cost minimization decisional tool for this manufacturing process [21]. The tool selected the optimal set of disposable culture vessels for a user-specified annual demand, lot size, cell dose and user-specified manufacturing constraints, i.e. maximum allowed number of culture vessels per lot, which was set to 100. In its estimation process, the tool calculated the material (i.e. media, disposable culture vessels), labor, quality control and equipment costs involved in the expansion and differentiation stages of the process for a battery of sequential culture vessel combinations (see Fig. 2). Additional parameters utilized during the cost calculations were the overall yield of the manufacturing process and the expansion fold of the hES cells. In that way the upstream cost of goods were estimated.

The cost minimization decisional tool did not include the downstream component of the manufacturing process (e.g., finishing, packaging, shipping), therefore 95% credible ranges were derived for cost estimates in four different settings: two cell culture methods (adherent and suspension), and each with two supply levels (50 and 500 doses per year) (also see Table 1). The credible ranges were used to fit Gamma distributions, i.e. the lower and upper bounds of the credible ranges for every cost estimate were equated to the values of cumulative density functions (CDF) at values of CDF = 0.025 and CDF = 0.975. The distributions were then used directly in the health economic modeling software.

We simulated four manufacturing modes: local production (e.g. at one University only), large scale production (one central lab produces all the doses and then ships them to the hospitals), and two scale-out production modes (local and large scale). The scale-out scenarios involved a network of several labs producing their own doses at their respective location but collaborating with each other through sharing expertise and research resources. We simulated one scale-out scenario for local productions and one for large scale productions. The local and large scale production scenarios assume a demand of 50 and 500 doses per year respectively. In general, the scale out approach may engage the capabilities of multiple local institutions and companies. It could, however, also contribute to unequal product quality and an increased overhead costs.

We estimated the long-term capacity to perform device implants in Canada to be 10 clinical centers. That estimate was derived by counting the hospitals on the list of transplant centers by the Canadian Organ Replacement Register in which clinicians performed islet cell transplants or other transplants of at least three different kinds of organs [35, 36]. We took this as clinical capacity to carry out transplantations of beta cell replacement devices that do not require immunosuppression. In the short term there could be two centers, one for Western Canada and one Eastern Canada.

We describe the demand for and composition of the doses of beta cell replacement tissue as follows. The annual demand of beta cell replacement doses was based on the current number of islet cell transplants in Canada and assumed to be 50 per transplant center, which was derived as linear extrapolation of transplant numbers in at the University of Alberta Hospital. Further we presumed the number of lots produced per year is 10, i.e. about one per month, and a minimum of 500 million cells are required per dose. Those numbers were derived from considerations of cell quality loss over time and the production figures above. Based on experience in the biotechnology sector the production assumed one of two production technologies, adherent or suspension cell culture approach, each with optimized production set ups for the two demand options (50 or 500 doses per year).

As a substantial simplification due to the novelty of the membrane technology, we presumed the cost of the device casing without the cells is off-set by reductions in costs through increased ability to plan transplantation times and processes.

Additional challenges of the donor-harvested approach in Canada could lead there to more readiness to adopt a stem cell based therapy approach even with initially higher costs. Among those challenges one needs to consider the relative shortage of organ donations, combined with great geographic distances between donors and the islet processing and transplantation site [44]. These factors can lead to two kinds of costs of timely organ transport. The monetary costs are sometimes covered by different regional health care services or air lines. Air transport companies are known to occasionally ship donor organs free of charge. Nevertheless non-monetary costs are unavoidable. An example is the cost of organ deterioration from with progressive cold ischemia can mar graft yield later on.

All those costs tend to be less for more densely populated countries, or even regions with different organ donation legislation, which can make a considerable difference in donor availability [45]. International coordination of donor organ availability could also further increase the efficient use available clinical resources. However, such coordination tends to come with substantial political and practical complexities, which require further research but are beyond the topic of this study.

While high volume implant production can theoretically be cheaper one needs to weigh that with several considerations regarding demand, clinical capacity and other practical limitations. One of those considerations is that stem cell-derived doses are currently as perishable as the donor derived cells. This means they have to be used within about 12–36 h of completion of the production process.

The difference between stem cell derived tissue and harvested cells is here that one can determine the time when the tissue is ready. Instead of the cell dose coming into the hospital more or less randomly at any time of the day or night, one can time the production process so that the dose or doses arrive at the hospital at a predefined day and time of the day.

In that way one can avoid the additional costs involved in nightly or short notice transplantations. But graft doses still have to be transplanted as quickly as possible. If several doses arrive at the same time it is also the case that all need to be transplanted within a short period of time. That could be accomplished if for example every week or every month 10 doses arrive at a hospital and are then all transplanted into 10 patients within the same day.

For that reason the number of lots produced per year is important. Every time a lot is produced all the doses of the lot have to be used within about one day or else go waste. That is because currently it is not possible to preserve beta cell progenitors over long periods, e.g. via cryo-preservation. In this context, it is advantageous from an economic perspective to produce several lots per year with smaller lot sizes, since it is impossible to transplant for example 500 doses in one day – even if spread over 10 transplantation centers.

Given the nature of the cells, transporting the patients to a central location – as is done currently - might be a better idea than transporting the cells to multiple patient hospitals across Canada. That is because transport of patients may actually be more affordable than the sum of: a) the health lost through the certain quality loss in the highly perishable cells through transport duration, b) the monetary costs from transporting the cells on a punctual just-in-time basis, c) the costs of duplication of in-hospital infrastructure and staff training.

Ongoing breakthroughs for example in current good manufacturing practice (cGMP) and mass cell expansion and limit the longevity of our estimates and modeling efforts. Breakthroughs include the genetic engineering technique CRISPR (Clustered Regularly Interspaced Short Palindromic Repeat) and the use of modern bioreactors which aid various kinds of bioprocessing [46, 47]. All those technologies have great influence on the capacities of researchers to generate new or more affordable ways of producing transplantable tissue.

We acknowledge that intensive insulin therapy as only comparator strategy to stem cell-based beta cell replacement therapy did limit the scope of our results. However, we consider the comparator and hypothetical patient cohort in our model to be appropriate because of the patient population under consideration. We explicitly limit our study population to patients who: 1) do not have the degree of major comorbidities which would justify risks of whole organ transplantation, 2) are on intensive insulin treatment and 3) are candidates for islet transplantation. Still, we think further studies in a North American context do need to include donor-base islet transplantation as one of the comparators. Other donor-based approaches to address type 1 diabetes could also be integrated.

The cost of the semi-permeable membrane in which the cells are enclosed had to be estimated doe to lack of data. This and the fact that we did not change the follow up costs and frequency compared to the study on islet transplantation are clear limitations of this study. However, in light of the technology under consideration being new and containing living cell tissue, the differences between the actual and our estimated follow-up costs are likely smaller than for a less complex implant device.

We expect that implantation would likely be an outpatient procedure with much more limited risks compared to islet transplantation, or even whole organ transplantation. In this study we presumed the new treatment technology would still be require an in-patient procedure including four days of hospital stay. Compared to that estimate, an out-patient procedure would further reduce the costs of stem cell-based beta cell replacement therapy while increasing patient quality of life.

One of the main goals of using re-extractable sheets for transplantation, instead of the standard cell injection into the liver, is to shield the cells from being attacked by the immune system. Since this is an early health technology assessment of a very new technology we made the assumption that this goal can be achieved without the use of imunosuppressive drugs. If future developments show that this is not the case then immunosuppression would be necessary and with it would come the usual costs and side effects as mentioned elsewhere [20].