Translatability score revisited: differentiation for distinct disease areas

FDA market approvals for the years 2012–2016 were analyzed to define the importance of the score items for the following six disease areas: cardiovascular, oncology, psychiatric, anti-viral, anti-bacterial/fungal and monogenetic orphan diseases (Additional file 1: Tables S1–S6).

During that time span oncology approvals were most frequent (46%, Fig. 1a), while in psychiatrics only 4% of the total number were approved (Fig. 1a). This is in line with a relatively low likelihood of approval of 6.2% in psychiatrics determined for earlier years (2006–2015) [11]. Oncology drugs had a twofold higher rate of first cycle approval than psychiatric drugs, which had the lowest first cycle review approvals [11]. For infectious diseases the likelihood of approval is high with 19.1% [11], while numbers of approval in our sample (anti-viral: 9%, anti-bacterial/fungal: 10%) are lower than in cardiovascular (16%), and monogenetic orphan diseases (15%, Fig. 1a). The number of approved drugs depends on the number of drugs in the pipeline which is highest for oncology [12], therefore resulting in high approval numbers despite the low likelihood of approval (5.1% [11]). The astonishingly low likelihood of approval for oncology could be partially due to commercial decisions and portfolio prioritization [11, 13].

Oncology has the highest percentage of total US approvals in every analyzed year (Fig. 1b, primary y-axis, continuous lines). For the monogenetic orphans there is a trend towards more approvals (Fig. 1b, primary y-axis, continuous lines), which may be due to the intense support of the development of drugs against orphan diseases by the FDA.

Analyzing the different years there is a peak of total approvals in 2015, with a remarkable decrease in 2016 (Fig. 1b, secondary y-axis, broken line). In total (including all areas of disease) there have been only 22 approvals in 2016 compared to 30 in average in the years 2007–2015 [14]. The same trend was true for approvals by the EMA [15]. This may be due to the fact that some of the drugs which have been approved in 2015 have been originally scheduled for 2016, and for some, which have been scheduled to 2016 approval has been delayed, due to more complete response letters issued by the FDA. These letters were often addressing deficiencies in good manufacturing practice [16].

Companion diagnostics is not a completely new phenomenon though very fashionable today. The use of tests to guide the effective and safe use of drugs is an essential part in the process of patient care and has been performed for a long time. Examples are glucose-6-phosphate-dehydrogenase deficiency testing and Rasburicase [15], or the identification of defective clotting factors for the diagnosis and treatment of patients with coagulation disorders. CDx was widely recognized only with the success of trastuzumab and imatinib [17].

There is no consistent definition of CDx [17]. For our definition see “Methods”. The definition of the FDA for CDx is stricter: A companion diagnostic is a medical device, often an in vitro device, which provides information that is essential for the safe and effective use of a corresponding drug or biological product, including identification of the right patient and monitoring response. The use of a companion diagnostic device with a particular therapeutic product is stipulated in the instructions for use in the labeling of both the diagnostic device and the corresponding therapeutic product [18].

In our extended definition, all assays mentioned in the FDA package insert dealing with efficacy, safety and/or monitoring were considered to be CDx, even if not explicitly mentioned and not found to be mandatorily tested by an FDA approved device. CDx have been found in all disease areas except psychiatrics (Fig. 2, column 1) as alluded to above. In our sample CDx stipulated mandatory in the FDA package insert were only present in oncology (Fig. 2, column 2). Our definition results in a high number of CDx for anti-infectives and monogenetic orphans, as susceptibility, hypersensitivity, resistance testing or underlying genetic cause should be tested for prior to therapy. The fraction of CDx is particular high for anti-viral drugs, as most of the approved drugs in the evaluated time period have been drugs against HIV or HCV (Additional file 1: Table S4), requiring resistance testing or genotyping, respectively. Therefore all tests for the anti-virals are genetic assays (Fig. 2, column 3). In contrast, there were no genetic tests for anti-bacterial/fungal drugs, as the recommended or necessary tests are dealing with susceptibility (antibiograms). For cardiovascular, oncology and monogenetic orphans, genetic tests are needed in some cases (Fig. 2, column 3) but also other tests, including enzymatic tests (Additional file 1: Tables S1, S2, S6), may be recommended. The high number of CDx in monogenetic orphans is due to the fact that the underlying genetic cause has to be analyzed using genetic or enzymatic tests to get a clear diagnosis.

In order to develop CDx, good understanding of the underlying mechanisms for the disease is important. If this is the case the likelihood for success is much higher so that it is not astonishing that the probability of successful transition from phase I to approval is 25.9% for drugs coming with a CDx test compared to 8.4% for drugs without predictive biomarker [11]. As evident in Fig. 2, the lack of CDx for psychiatrics is in line with the lack of biomarkers as a general feature of this disease area.

CDx may further facilitate the clinical studies as a smaller number of patients may be needed to show an effect, saving resources and time spent on clinical development [17]. The use of a biomarker in phase III increases the success to 62% compared to 28% overall success rate in phase III [11].

The potential of CDx in development will be mainly reflected in biomarker-related items of the scoring approach [5].

Animal models are important to provide data on safety and efficacy (so-called animal-PoC, proof of concept) at the preclinical level; yet, there are large differences regarding availability and usefulness of animal models between disease areas. The average number of animal models reported for efficacy varied considerably between disease areas ranging from 0.4 to 8.4 (Fig. 3a). While relatively few different animal models could be found for oncology, cardiovascular, anti-viral and monogenetic orphan drugs, larger numbers of models have been reported for psychiatrics and anti-bacterial/fungal drugs. Despite the multitude of animal models used in psychiatrics, the average number of positive results translating into human findings was lowest in this area (Fig. 3b), indicating a low predictivity of animal models into clinical trials. In the other areas the contribution of animal models to translation was higher, as indicated by a greater rate of successful translation (Fig. 3b). The small number animal models tested for anti-viral drugs seems to reflect the fact that most of the approved drugs are treating HIV or HCV (Additional file 1: Table S4), which are highly specific for humans, and, therefore, animal models are not required by the FDA [19, 20]. The large number of animal models used to develop anti-bacterial/fungal drugs reflects the need for testing many different infection models including those on different sites of infection and sepsis (Additional file 1: Table S5).

Though not particularly prevalent in the sample of drug approvals analysed here, the weight factor for animal evidence should be increased (doubled or even tripled) case by case if primate studies were performed; they however cannot replace the value of human studies though their predictivity is certainly much higher than that of e.g. rodent studies.

Based on the evaluation of the FDA reviews and related literature the translatability scoring system published 2009 [4] has been modified for different disease areas with particular emphasis on biomarkers and animal models as described above. Changes of weight factors were mainly driven by these aspects with weight changes for other aspects being adapted oppositely to keep the sum of weight factors at 100%. Changes in the weights differing by more than two points from the original are highlighted in italic (Table 1).

Changes in the weights of the score have been performed according to the results of the FDA reviews and package inserts. Important weights were doubled (model compounds, clinical trials for psychiatrics), tripled (disease sub-classification and responder concentration, or quadrublicated (genetics for monogenetic orphans). Weights for points which are less important have been reduced. As the overall score needs to be 100 all other scores have been adapted accordingly.

The weights for cardiovascular diseases have been left unchanged, as the original score has been developed against a cardiovascular background. For oncology the weight of animal models has been increased, as our evaluation of the FDA reviews showed a good correlation between animal and clinical data (Fig. 3 and Additional file 1: Table S2). It is however of concern that a publication bias might exist for animal models as possibly not all negative results may have been discussed in the FDA documents. The weight for personalized medicine aspects and biomarkers have been increased in oncology in reflection of a large prevalence of CDx (45.1% of all oncology drugs, 23.5% stipulated as being mandatory by the FDA) this points to the high relevance of personalized medicine in oncology [21].

The weights for the starting evidence in psychiatric diseases have been reduced as reliable animal models and in vitro testing hardly exist (Fig. 3, Additional file 1: Table S3 and [22, 23]). The weights for model compounds which already made it into human application and clinical trials have been increased; this reflects the lack of animal models and suitable biomarkers rendering human evidence particularly important. Biomarkers for safety and efficacy are very difficult to develop in psychiatric diseases [10] and accordingly the weights of the related items (starting evidence, biomarkers for efficacy and safety prediction, biomarker strategy and personalized medicine) have been lowered. Surrogate or endpoint strategy scoring has been doubled, as the lack of animal models renders the surrogate or endpoint strategy especially important. The weights for personalized medicine aspects have been lowered since CDx are not found in this category (Fig. 2, Additional file 1: Table S3).

For virus-borne diseases the weight for in vitro data has been increased, while the weight of animal models has been decreased. Many viruses are highly specific for humans (for example HIV and HCV) and therefore reliable animal models are lacking. For example for HIV several animal models are available but none of them adequately reflects the human HIV [24]. This results in a greater importance of in vitro testing in this field. Further in vitro testing is well established for anti-viral (and anti-microbial) drugs [25]. Additionally the aspect of personalized medicine is quite important in this field e.g. HIV- or HCV-mediated diseases require special testing for resistance or genotyping, respectively (Additional file 1: Table S4).

For bacteria- and fungi-mediated diseases several reliable animal models exist (Fig. 3, Additional file 1: Table S5 and [26]), so that the weight for this aspect is higher than for anti-viral drugs. The aspect of personalized medicine is also quite important in this field due to susceptibility testing [27] and therefore the weight has been increased for this aspect.

For orphan genetic diseases animal models often exist (Fig. 3, Additional file 1: Table S6) as they do in cardiovascular diseases, so that the related weight has been left unchanged. The weight of human evidence, especially genetics, has been increased as by definition monogenetic orphan diseases always have a known genetic cause; this knowledge facilitates the translational process. The weight for model compounds has been lowered, since orphan drugs are mostly first in class drugs. Further the score for personalized medicine has been increased, as specific testing for the underlying genetic disease is performed in most cases.