Background
Methods
Research question
Search
Additional search strategies
Selection of papers
Selection of data
Human test | ||
---|---|---|
Positive (HP) | Negative (HN) | |
Animal test | ||
Positive (AP) | TP | FP |
Negative (AN) | FN | TN |
Sensitivity = TP/HP | ||
Specificity = TN/HN | ||
Positive predictive value (PPV) = TP/AP | ||
Negative predictive value (NPV) = TN/AN | ||
Accuracy = (TP + TN)/(TP + FP + FN + TN) |
Analyses of translational success rates
Type of analysis | Type of definition for translation | Used value for % translational success |
---|---|---|
Correlation | Continuous | Squared correlation coefficient (r2) expressed as percentage |
Regression | Continuous | Squared correlation coefficient (r2) expressed as percentage |
Fold error | Binary | Percentage below twofold error |
Meta-analysis | Continuous | Percentage overlap of 95% CIs |
Binary analyses | Binary | % accurate or PPV or NPV |
Risk of bias and reporting quality
Results
Search and selection
Characteristics of the included papers
Studies addressing translational success rates
Studies addressing general medical sciences and efficacy
Study ID | Field of research | Summary of findings |
---|---|---|
Briassoulis_2014 | Sepsis | Animal studies show clear protective effects of HSP72 in sepsis, human studies are inconclusive |
Brossi_2015 | Orthopedia | Equine studies on the efficacy of platelet rich plasma (k = 63) mostly show positive results, human studies (k = 60) have variable outcomes. Beneficial results are more frequent in studies with a high risk of bias |
Contopoulos-Ioannidis_2003 | Diverse | Out of 64 publications of animal studies in highly cited basic science journals, 16 interventions were tested in a published clinical trial, 12 of which had positive results |
Corpet_2005 | Oncology | Relative Risks after treatment were discordant for 2 out of 11 compounds between rats and humans, and mice and humans |
Faggion_2009A | Dentistry | pocket depth reduction and attachment level gain were similar for animals and humans |
Hackam_2006 | Diverse | Successful translation is not predicted by study methodology, but it is predicted by the presence of dose–response gradients in animals |
Johnson_2001 | Oncology | Xenograft models that were available at this stage could not reliably predict the clinical response |
Lindl_2005 | Diverse | The publications resulting from 51 animal ethics approvals were followed. 16 projects were relevant to humans and resulted in 63 publications that were cited 1183 times. 97 citations were clinically oriented, of which only 4 evidenced an animal-human correlation. The hypotheses verified in animals failed in every respect in humans |
Perel_2006 | Diverse | For 6 interventions, animal and clinical studies were concordant for 3 and discordant for the other 3 |
Steinberg_1987 | Pancreatitis | With the same 5 interventions, 81% of animal studies had a positive outcome, and only 7.7% of the human studies. |
Sultan_2017 | Cardiology | Most of the human data did not show any effect of cannabidiol, while the animal studies did |
Valles_2018 | Dentistry | Results from animal and human studies are concordant |
Voskoglou-Nomikos_2003 | Cancer | None of the primary analyses showed a significant correlation |
Whiteside_2008 | Pain | For effective pain treatments, the correlation between human and rat effective doses is good |
Yardley_2016 | Alcohol abuse | Out of 49 animal studies (on 8 drugs), 45 showed positive results. Out of 76 human studies, 56 showed positive results. |
Yen_2014 | Dentistry | Animal models and human results showed similar bone filling ratios |
Studies analysing adverse events and toxicology
Study ID | Field of research | Summary of findings |
---|---|---|
Alden_2011 | Carcinogenicity | Out of 287 registered drugs that were tested in rats and mice for carcinogenicity, results were concordant with humans for 146 |
Allen_1988 | Carcinogenicity | Correlation of carcinogenic dose between animals and humans ranged from 0.49 to 0.90 depending on the analysis |
Bailey_2013 | Safety | All likelihood ratios (LRs) are larger than 1, indicating predictive value of the experiments in dogs. Inverse negative LRs (iNLRs) are very small, indicating relatively limited predictive value of negative results in dogs for humans. Positive LRs (PLRs) for dogs are large; if toxicity is observed in dogs, it is likely to occur also in humans. There is no correlation between positive predictive values (PVVs) and PLRs |
Bailey_2014 | Safety | All LRs are larger than 1, indicating predictive value of the experiments in rats, mice and rabbits. iNLRs are very small, indicating relatively limited predictive value of negative results in these species for humans. PLRs for these species are large; if toxicity is observed in rats, mice or rabbits, it is likely to occur also in humans. Both PLR and iNLR depend on sample size |
Bailey_2015 | Safety | All LRs are larger than 1, indicating predictive value of the animal experiments. iNLRs are very small, indicating relatively limited predictive value of negative results in animals for humans. PLR for non-human primates (NHPs) is large; if toxicity is observed in NHP, it is likely to occur also in humans |
Brown_1983 | Teratogenicity | Correct positives: 30–97%; correct negatives: 35–80%; animal to human lowest effective dose ratio: 1.8–50 |
Claude_2007 | Adverse events | 70% of human adverse events was predicted by animal models. Predictivity is higher for non-rodents than rodents. Predictivity was highest for haematological and cardiovascular, and lowest for cutaneous and ophthalmological adverse events |
Crouch_1979 | Carcinogenicity | Data for carcinogenic potency correlated |
Davis_1998 | QT prolongation | Out of 9 noncardiac drugs that show QT prolongation in humans, literature on dog cardiac effects was found for 7; 6 showing QT prolongation, 1 showing increased mortality |
Ennever_2003 | Carcinogenicity | Sensitivity appears to be high, but the lifetime rodent bioassay lacks accuracy. Sensitivity decreases if only results that are positive in both rats and mice are considered positive. The LRB produces many false positives and false negatives |
Fletcher_1978 | Adverse events | Correlations between animal toxicity and human adverse events are considerably more frequent than discrepancies. Gastro-intestinal adverse events show the best correlation |
Fourches_2010A | Drug-induced liver injury | The concordance of liver effects between rodents and humans (44%) and between non-rodent species and humans (40%) was low |
Freireich_1966 | Toxic dose | Results in preclinical tests correlate remarkably well with results in man |
Goodman_1991 | Carcinogenicity | For 18 out of 20 examined chemicals with sufficient evidence, human and rodent evidence are consistent |
Hoffmann_2018 | Skin sensitization | Overall accuracy in skin sensitization prediction from animal to human was 74%, which decreased to 45% when considering five categories of potency |
Igarashi_1995 | Adverse events | Out of 31 pharmacological items tested after systemic administration, 17 showed a significant association with any clinical adverse reaction |
Litchfield_1961A | Adverse events | 18 out of the 53 physical signs observed in man were predicted correctly in rats; 29 out of the 53 in dogs |
Litchfield_1962 | Adverse events | Out of the 86 physical signs analysed in animals, 64 accurately reflected occurrence or absence in man |
Monticello_2017 | Adverse events | Excluding subjective adverse events, for rodents, PVV ranged from 0 to 54% and NPV ranged from 69 to 96%; for dogs, PVV ranged from 0 to 52% and NPV ranged from 76 to 96%; and for monkeys, PVV ranged from 0 to 91% and NPV ranged from 70 to 100% |
Olson_2000A | Adverse events | In any species tested, 71% of human adverse events was predicted. Predictivity is higher for non-rodents than rodents. Predictivity was highest for haematological, cardiovascular and gastrointestinal toxicities, and lowest for cutaneous toxicities |
Schein_1970 | Adverse events | For the prediction of certain adverse event in humans, administration of highly toxic dose levels to animals is needed |
Schein_1973a | Adverse events | For most organ systems, combining dog and monkey data reduces false negatives for prediction of human adverse events for anticancer drugs |
Schein_1973b | Adverse events | Correct predictions of anticancer drug-induced adverse events are accompanied by a high percentage of false positives |
Schein_1975 | Adverse events | Results from 13 additional drugs generally overlap with the preceding analysis |
Tamaki_2013 | Adverse events | 37% of adverse drug reactions in humans were predicted from animal studies |
VanMeer_2012 | Severe adverse reactions | Performed animal studies are not sensitive enough to predict post-marketing serious adverse reactions |
Weaver_2003 | Adverse events | No significant associations were observed between human and guinea pig data |
Wilbourn_1986 | Carcinogenicity | Sensitivity for the predictivity of animals for human carcinogenicity is high (84%), and there is good consistency between animals and humans in target organs |
Studies addressing pharmacokinetics
Study ID | Field of research | Summary of findings |
---|---|---|
Akabane_2010A | Absolute bioavailability | Bioavailability in cynomolgus monkeys is unsuitable for predicting PK in humans |
Bachmann_1989 | Clearance | Predicted values are in the same order of magnitude as actual values |
Bachmann_1996A | Volume of distribution | Human volume of distribution and half-life values can be predicted from those in rats |
Boxenbaum_1982A | Clearance | It is not possible to reasonably predict human pharmacokinetic parameters from knowledge of these parameters in dogs |
Caldwell_2004A | Clearance | There is a reasonable correlation between human and rat clearance and half-life; and a good correlation for volume of distribution, but only 52–65% of drugs show < twofold error. Go/no go decisions based on only rat data should be avoided |
Campbell_1994A | Clearance | Predictive accuracy for clearance from rat, dog and monkey is acceptable. The dog is a poorer predictor of clearance than the rat |
Cao_2006A | Oral bioavailability | Oral bioavailability does not correlate between rats and humans; R2 = 0.29 while intestinal permeability correlates better; R2 = 0.70 |
Cheng_2008 | Oral absorption | Human intestinal absorption cannot be precisely predicted by a single screening assay |
Chiou_1998A | Oral bioavailability | Oral bioavailability correlates between rats and humans, and to some extent between dogs and humans |
Chiou_2000a | Oral absorption | Similar gastrointestinal absorption may be obtained when doses in humans (/kg body weight) are 5–7 times lower than in rats |
Chiou_2000bA | Oral absorption | R2 = 0.51–0.90 for oral absorption between dogs and humans; plasma level peak times seem to be shorter for dogs. R2 = 0.95 for oral absorption between rats and humans. |
Chiou_2002A | Oral absorption | Oral absorption correlates well between monkeys and humans: R2 = 0.97; bioavailability correlates to some extent between monkeys and humans: R2 = 0.50; clearance correlates between monkeys and humans: R2 = 0.82; time to peak concentration was generally similar in humans and monkeys |
DeBuck_2007A | Volume of distribution | Predictions of plasma concentrations after oral dosing are reasonable. Prediction of volume of distribution improves when accounting for interspecies differences in plasma protein binding. 18 out of 19 drugs had a predicted half-life within twofold of the actual observed half-life |
Dong_2011A | Volume of distribution | For Monoclonal antibodies with non-linear kinetics, prediction is poor, with up to 6.3-fold differences |
Evans_2006 | Clearance, distribution volume and residence time | Percentages of correct predictions for clearance, distribution volume and residence time for rat, dog and monkey varied from 29 to 91%, and the average margin of error from 44 to 159%. The authors note that the outcomes are different from similar analyses of other compound datasets |
Fagerholm_1996 | Jejunal permeability | For passively absorbed compounds (n = 8), the correlation is high; R2 = 1.0. For passively absorbed compounds, rat permeability estimates can be used to predict human oral absorption |
Fagerholm_2007a | Fraction excreted unchanged | Out of 25 compounds, 11 had a fraction of 0 excreted unchanged in both humans and rats. For 9 out of 14 compounds with renal excretion in rats and humans the major route of elimination differed between species. Findings for monkey–human comparisons were roughly comparable |
Fagerholm_2007b | Unbound fraction in plasma | The fraction unbound in plasma correlates between rats and humans; R2 = 0.67. Different prediction methods show different accuracies |
Goteti_2010A | Clearance | Two-species scaling can be useful, but the prediction of clearance from ≥ 3 species is more accurate |
Grime_2013A | Clearance | For 19 out of 22 drugs, rat unbound biliary clearance exceeded human clearance by factors ranging from 9- to 2500-fold. Human–dog differences in biliary clearance were less dramatic than human-rat differences |
He_1998A | Oral bioavailability | In human and rat there is generally a good correlation for oral bioavailability, in human and dog there is no apparent correlation. |
Hosea_2009A | Clearance | Single species scaling is as accurate or more accurate than multiple-species allometry |
Ito_2005A | Intrinsic clearance | Human clearance is better predicted by modelling based on in vitro microsomal data than on animal data |
Jolivette_2005A | Clearance, volume of distribution | Molecular properties may be used to improve extrapolation from animal to human clearance |
Jones_2012A | Clearance, mean residence time | Prediction was within twofold for 5 out of 7 compounds |
Jones_2016C | Intestinal availability | There is little evidence that one animal species is sufficiently predictive of human first-pass metabolism to be used as a stand-alone model |
Kalvass_2007A | In vivo potency (EC50), clearance | In vivo mouse brain half-lives are almost identical to human half-lives. In vivo preclinical to clinical extrapolations are superior to extrapolations from in vitro tests |
Lave_1999 | Clearance | Human clearance is most accurately predicted from a combination of in vivo animal and in vitro animal and human data |
Lave_2002 | Clearance | Predictions based only on in vitro data are at least as accurate as those based on multiple species data |
Lennernas_2007 | Jejunal permeability | A rat model can be used to predict oral drug absorption, but not drug metabolism or oral bioavailability |
Ling_2009 | Clearance | Human clearance might be accurately predicted from monkey data |
Mahmood_1996a | Clearance | Human clearance can be estimated from animal data, but caution and scientific judgement are needed for interpretation |
Mahmood_1996bA | Clearance, volume of distribution | A new approach incorporating brain weight in the model improves prediction of clearance |
Mahmood_1996cA | Clearance, volume of distribution | Three or more species are needed for reliable prediction of clearance, while volume of distribution is predicted equally well using data from two species or more |
Mahmood_1998a | Clearance | Mean residence time can be predicted reasonably well for man and can be used for prediction of half-life |
Mahmood_1998bA | Clearance, volume of distribution | Caution should be employed when interpreting clearance predictions for renally excreted drugs. Predicted volumes (error − 65.6% to 139.4%) and half-lives (error − 41.8% to 100%) were comparable with observed values in man. |
Mahmood_1999 | Selection of 1st in human dose | The half-life and bodyweight correlate poorly; body weight is not useful as a predictor |
Mahmood_2000a | Bioavailability | All tested approaches predicting human bioavailability from animal data are inaccurate |
Mahmood_2000b | Protein binding | Unbound human clearance cannot be predicted any better than total human clearance from animal data |
Mahmood_2001 | Maximum tolerated dose | Maximum tolerated dose can be predicted with reasonable accuracy using interspecies scaling |
Mahmood_2003 | Selection of 1st in human dose | Animal PK data from a dose not producing adverse events can be used to estimate a suitable human starting dose |
Mahmood_2004 | Clearance | More than two species are needed for reliable clearance predictions of protein drugs |
Mahmood_2006 | Clearance | There is no single method for predicting human clearance from animal data for all classes of drugs |
Mahmood_2009 | Clearance | Predictions based on at least 3 animal species remain more accurate than one or two-species methods |
Mahmood_2012 | Clearance, volume of distribution | The human clearance of drugs that are excreted in the bile can be predicted with reasonable accuracy from animal data. The volume of distribution does not appear to be affected by biliary excretion |
Mahmood_2013 | Concentration–time profiles | Human concentration–time profiles of therapeutic proteins can be predicted reasonably accurate from animal data |
Mahmood_2013 | Clearance, volume of distribution | Concentration–time profiles are accurately predicted for most time points |
Mahmood_2016 | Clearance, volume of distribution | Human plasma time–concentration profiles, oral clearance and volume of distribution can be predicted with reasonable accuracy |
McGinnity_2007 | Clinical dose, maximum concentration & volume of distribution | There is a reasonable to good correlation between projected and clinical human dose, observed and predicted maximum concentration for a given human dose and predicted and observed human volume of distribution |
Musther_2014 | Oral bioavailability | Bioavailability in animals is not quantitatively predictive of bioavailability in humans |
Nagilla_2004 | Clearance | Prospective allometric scaling is a suboptimal technique for estimating human clearance data from in vivo preclinical data |
Naritomi_2001 | Clearance | Animal data improve predictions of human clearance from in vitro liver microsomes |
Obach_1997 | Volume of distribution, clearance | Methods for accurate prediction of human PKPD based on animal data do not currently exist, but many methods result in adequate predictions |
Paine_2011 | Clearance | The most accurate predictions of human renal clearance are obtained from a direct correlation with dog renal clearance. Adding data from rats decreases predictability |
Pogessi_2004 | Volume of distribution, clearance | In most cases, animal-based predictions are within two or threefold of those observed in humans |
Rocchetti_2007 | Active dose | Therapeutically active concentrations of anticancer drugs can be estimated from preclinical studies |
Sanoh_2012 | Clearance | PXB chimeric mice can be used for at least semi-quantitative prediction of human clearance and half life |
Sanoh_2014 | Metabolism | Human metabolites were sufficiently predicted from the animal data for 10 out of 16 compounds; predictions were insufficient for 6 out of 16 compounds |
Sawada_1985A | Clearance, volume of distribution | Predictions for human clearance, volume of distribution and half-life from rat data were successful for most drugs, with marked exceptions |
Sawada_1985B | Volume of distribution | Prediction of human volume of distribution based on animal plasma unbound fraction was successful for most drugs |
Schneider_1999 | Clearance | Dog and rat in vivo hepatic drug clearance data appear unrelated with human data |
Sietsema_1989 | Oral bioavailability | Absolute bioavailability does not correlate well between species |
Takahashi_2009 | Bioavailability | The bioavailability in cynomolgus monkeys was low compared to that in humans for most drugs tested |
Tang_2005 | Clearance | A new mathematical model based on unbound fractions can improve prediction of human clearance from animal data |
Tang_2006 | Clearance | There is no strong evidence that human systemic clearance is more predictable from animal data than human oral clearance |
Wajima_2002 | Clearance | Multiple linear regression of animal data generally predicts human clearance better than allometric methods |
Wajima_2003 | Oral clearance | The partial least square method based on animal data generally predicts human oral clearance better than allometric approaches |
Walton_2004 | Clearance | Average differences in the internal doses between humans and animals were 1.6 for dogs, 3.3 for rabbits, 5.2 for rats and 13.0 for mice |
Wang_2010 | Clearance | Human clearance can generally be predicted well from animal data with simple allometric scaling |
Ward_2004a | Clearance | Generating data in multiple non-human species does not always result in improved prediction |
Ward_2004b | Volume of distribution | The monkey provides the most accurate PKPD predictions for humans. The allometric exponent cannot be used as a reliable marker of predictive success |
Ward_2005 | Clearance | The rat is not as accurate a predictor as the monkey, but still affords reasonable human predictivity |
Ward_2005 | Oral systemic exposure | Liver-corrected oral exposure was within twofold of human for 30% of compounds for rats, and for 48% for dogs. The monkey was the preclinical species most similar to humans |
Ward_2008 | Clearance | Reasonable predictive accuracy of pharmacokinetic parameters in humans can be achieved with African green monkeys |
Ward_2009 | Bioavailability | The African green monkey provides similar predictivity for human oral exposure as other monkeys |
Whiteside_2010 | Maximum concentration | Rat models for pain predict effective exposure levels in humans. Effective plasma concentrations also correspond. |
Wong_2004 | Clearance | The chimpanzee serves as a valuable surrogate model for human pharmacokinetics |