By Dr. Daniel Calcinaro, PhD Graduate at ZLC and Dr. Mustafa Çagri Gürbüz, ZLC Professor.
Clinical trials are perhaps the most critical stage in bringing a new pharmaceutical product to market, but historically the costs of running trials were seen as a fairly marginal component of R&D expenditure. That is emphatically no longer the case.
A new drug can take 10-15 years to move from lab to pharmacy, a process costing typically €1 bn to €5 bn. Usually, there are three phases of pre-market trials (in each case comparing the new formulation with a placebo or an existing treatment). Phase One tests the drug on 50-100 healthy individuals for toxicity; Phase Two will explore the optimal way to administer the therapy across several hundred patients; and Phase Three, if the drug gets that far, tests safety and efficacy across several thousand patients.
Trial costs have increased greatly in recent decades, because of ever more rigorous regulation but also because trials are necessarily becoming global in nature. Effective trials require a sufficient pool of volunteer participants, that allows fast recruitment and enables results to be safely generalised against differing patient populations, ethnicities, and approaches to the management of the disease in question.
The pharmaceutical under trial has to be readily available as and when a trial participant enrols, and going forward if a multi-dose therapy is required. Unfortunately, it cannot be predicted when, nor where, a suitable patient will present, or how long the trial will last. To complicate things further, the drug will very likely have a limited shelf life, and the therapy may well involve a combination of drugs. Transnational trials have to be approved by multiple regulatory bodies. Increasingly, trials include emerging and less developed economies which may present infrastructure difficulties and associated costs with longer lead times and greater organisational challenges (ranging from varying import procedures to transport limitations).
By the time trials commence, the clock is usually already running down on intellectual property rights, so it is no surprise that any delay in trials can cost the drug company $1 million a day (not to mention the human cost of interrupted or abandoned therapies for trial participants, and the suffering or even death of the wider group that could potentially have benefited). Meanwhile the ‘pipeline’ of drugs in development is doubling every 11.5 years, the number of registered global trials increased by 24% just between 2021 and 2023, and the costs of supplying drugs to a trial, far from being negligible as may once have been the case, may now represent 20% of the drug’s final value, or 40% of the total cost of the trial.
Supply to clinical trials therefore represents a significant logistics challenge. Beyond the primary goal of saving lives, pharmaceutical companies are operating in a world of diminishing margins under pressure from health care providers so there is an imperative to reduce trial costs, for example by reducing overproduction and oversupply of drugs and particularly by enabling the trial to be completed more quickly, enabling the company to start earning a return, and creating savings which can be fed back into further R&D. And of course there is the drive for sustainability in all its forms, from reducing overproduction to the disposal of surplus or life-expired (and potentially environmentally hazardous) drugs.
In my thesis towards a PhD I have devised an approach to determining optimal production, inventory and distribution decisions for the supply of drugs to trials.
Simplistically, if patients could wait indefinitely, the exact number of doses required could be held in a central warehouse and downstream deliveries from a central warehouse postponed until we know where and when the patients have enrolled. That reduces inventory and delivery costs, but is not too good for the patients. At the other extreme, and if production and inventory costs were not an issue, each enrolment site could be supplied with enough doses to treat the entire trial, regardless of where participants enrol in practice. Obviously, neither of these extremes is realistic so this model is set up to follow the objectives of timely supply (being faster to market), reduction in oversupply, minimised cost, and saving lives and the planet, in that order although of course there are interconnections.
The model uses a multi-stage mixed integer linear stochastic program and, since this throws up a prohibitively large number of possible scenarios, a clustering scenario reduction methodology using a novel dissimilarity metric.
Among the ways in which this model differs from previous work is the ability to consider multiple production runs (reflecting that production of trial drugs may be at less than full scale, that drugs may be life-limited and that the duration of a trial can be uncertain, as well as the desire to prevent waste). Also, the model covers the possible effects of patients dropping out of trials as a result of poor supply chain decisions, meaning their dose isn’t available at the right time. Additionally, the model allows for a mix between supply to enrolment centres from local depots and supply direct from a central warehouse.
With the support of F. Hoffmann-La Roche Ltd., the model was successfully tested against three ‘real life’ clinical trials representing all phases of the trials process, very different disease types (diabetes, cancer, glaucoma) and covering different geographies (of from 2 to 6 countries, the latter involving 18 enrolment sites), arrival rates (over enrolment periods ranging between 132 and 547 days), shelf lives and number of doses per patient (from 1 to 15). Patients enrolment ranged between 45 for the Phase One trial to 190 in the Phase Three trial.
Importantly, the scenario reduction technique used reduced the computational challenge sufficiently that the model can be solved in a short enough time to be useful for managerial purposes. Significantly too, the scenario reduction technique employed appears to have minimal impact on the accuracy of the model.
There is scope to refine the model further. We could for example include disposal costs (of unused drugs) as an objective, considering the amount and timing of such disposals. We could extend to multi-product trials, and to treatments where the number of doses that will be required is uncertain. We could also incorporate updates of the expected rate at which patients will arrive in the trial as the trial progresses as this research shows that the costs are quite sensitive to forecasted demand, and there is scope to vary the model if different trial policies are adopted. But even as it stands, this model can be applied to most clinical trial situations.
For more information please contact Daniel Calcinaro, PhD ([email protected]) and Mustafa Çağrı Gürbüz, PhD ([email protected]).
