Professional Documents
Culture Documents
Radiation Oncology
biology physics
www.redjournal.org
Physics Contribution
Received Mar 12, 2019. Accepted for publication Jun 29, 2019.
Summary Purpose: A large proportion of preclinical or translational studies using radiation have
Many preclinical studies poor replicability. For a study involving radiation exposure to be replicable, interpret-
have poor replicability, able, and comparable, its experimental methodology must be well reported, particu-
partly because of undocu- larly in terms of irradiation protocol, including the amount, rate, quality, and
mented irradiation methods. geometry of radiation delivery. Here we perform the first large-scale literature review
We performed the first large- of the current state of reporting of essential experimental physics and dosimetry details
scale review of physics and in the scientific literature.
dosimetry reporting in radi- Methods and Materials: For 1758 peer-reviewed articles from 469 journals, we
ation biology studies, evalu- evaluated the reporting of basic experimental physics and dosimetry details recom-
ating the methods sections of mended by the authoritative National Institute of Standards and Technology sym-
1758 publications for 12 pa- posium.
rameters recommended by Results: We demonstrate that although some physics and dosimetry parameters,
the 2011 National Institute of such as dose, source type, and energy, are well reported, the majority are not.
Standards and Technology Furthermore, highly cited journals and articles are systematically more likely to
symposium expert panel. We be lacking experimental details related to the irradiation protocol.
Corresponding author: Yannick Poirier; E-mail: yannick.poirier@ Supplementary material for this article can be found at https://doi.org/
umm.edu 10.1016/j.ijrobp.2019.06.2545.
Disclosures: The authors have no conflict of interest to declare.
Int J Radiation Oncol Biol Phys, Vol. -, No. -, pp. 1e10, 2019
0360-3016/$ - see front matter Ó 2019 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/j.ijrobp.2019.06.2545
2 Draeger et al. International Journal of Radiation Oncology Biology Physics
demonstrate that although Conclusions: These findings show a crucial deficiency in the reporting of basic
some parameters, such as experimental details and severely affect the reproducibility and translatability of
dose, source type, and en- a large proportion of radiation biology studies. Ó 2019 Elsevier Inc. All rights
ergy, are well reported, the reserved.
majority are not. Further-
more, highly cited journals
are more likely to document
fewer parameters.
3542 Abstracts
749
Screened
Articles Excluded:
• No ionization radiation
experiment [413] 1035 2793 Articles
Abstracts Excluded:
• Imaging study only [271] Reviewed • Review articles [570]
• Not animals
• Not in English [169]
or cell cultures [240] 1758 Art.
• Not available
• Not external radiation [106] Scored (copyright, embargo,
• Other [5]
redacted) [10]
* Measurement geometry only scored if calibration standards protocol not specified; total points in dosimetry calibration cannot exceed 2.0.
details at all and 10 represents a description of all iden- was chosen to reduce the impact of outliers represented by
tified categories. Again, to be as conservative as possible, journals and articles with disproportionate impact factors
even incomplete, flawed, or contradictory reporting was and number of citations.
fully scored, as long as the required category was at least
partially described.
Publication demographics
Publication impact metrics We also analyzed the aggregate physics-reporting score for
various subcategories of publications. First, we divided by
The following publication impact metrics were recorded for radiation oncology, biology, or physics specialty journals.
each scored publication: journal impact factor, number of Although this distinction is somewhat arbitrary, specialist
citations, and number of citations per year. Number of ci- journals were identified as abbreviated journal names
tations was obtained from Google Scholar,36 and the containing “(Radiat) OR (Phys NOT (Biophys OR Phys-
number of citations per year was calculated simply by iol))”. This resulted in 389 publications from specialist
dividing the number of citations by the number of years journals and 1369 publications from generalist (at least in
between publication and 2018 (i.e. when the number of terms of radiation oncology, biology, or physics) journals.
citations were tabulated). The journal impact factor was Compared with papers from generalist journals, specialist
obtained from the SCImago Journal & Country Rank37 as journal publications tended to be in a lower impact factor
the number of citations per published article within the last range (average standard deviation: 2.2 0.8 vs 4.2
2 years, available for 2017 at the time of writing. 3.1) and to have fewer citations (30 39 vs 46 155) and
For 9 journals, distribution had ceased under the publi- citations per year (2.9 2.8 vs 4.6 12.4) but to be
cation’s original name. If the journal had changed names (2 similarly aged (10.4 5.8 vs 9.7 5.7 years since pub-
cases), the impact factor of the successor journal was re- lication). More importantly, the majority of specialist
ported instead. If the journal had ceased publishing entirely publications seem concentrated in a relatively narrow range
(7 cases), the most recent impact factor on record was used of impact factor journals, citations, and citations per year,
instead. Out of 469 journals, a single journal was unlisted whereas generalist publications occupy a much greater
on SCImago, such that there was no impact factor to report. range, as evidenced by their very wide standard deviations.
A value of 0.000 was used instead. We also separated publications by biological model,
Spearman correlation coefficients, along with associated yielding 961 in vitro studies, 736 in vivo (eg, whole animal)
significance (P value) and 95% confidence intervals, were studies, and 61 studies incorporating both, as well as by
calculated between the aggregate physics-reporting score source type (Fig. 2). An unpaired, 2-tailed, heteroscedastic
and each publication’s impact metricsdimpact factor, Student’s t test was performed between the aggregate
number of citations, and citations per year, as well as the physics-reporting score of specialist versus nonspecialist
number of years since publication. Because the Spearman journals, between in vivo versus in vitro studies, and be-
correlation coefficient is a nonparametric correlation coef- tween all different radiation sources. Type 1 errors were
ficient that strictly evaluates the monotonicity of the data, it avoided because significance values tended to be either
Volume - Number - 2019 20 years of incomplete physics in radiobiology 5
above 0.05 or very small (<0.001), such that false positives report x-rays or gamma rays without specifying the energy,
were very unlikely. manufacturer, or model. Finally, a relatively small propor-
tion of studies used more exotic radiation: a, b, or electrons
(3.8%), protons or heavy ions (6.3%), or other sources,
Results including neutrons (1.9%). The remaining studies (7.5%)
did not report the radiation source. In total, the source of
Our initial search identified 3542 articles from 968 journals radiation was unclear or ambiguous for 243 papers, or
published in the last 20 years (1997-2017), of which 1758 13.8% of the total, which loosely agrees with the findings
from 469 journals were admissible and evaluated. Figure 2 of Stone et al, wherein 30.1% of articles did not report the
shows the breakdown in biological subjects and radiation energy or radiation type.
sources in evaluated articles and shows that the majority of Figure 3 shows the breakdown of reported physics and
reported studies are performed in vitro or in murine (ie, dosimetric experimental details for each category. The
mice or rat) models, with all other models representing majority of authors reported the source (92.5%), the energy
6.8% cumulatively. Additionally, the majority of radiation (81.4%), and the manufacturer or model of the irradiator
biology studies continue to be performed with traditional (64.8%).
Co-60 and Cs-137 irradiators (approximately half of all The absolute dosimetry category was by far the least
reported sources), although they are increasingly becoming reported: Only 1.2% of all authors listed the protocol used
obsolete in both human patients and research. X-rays (kV for machine calibration, and only 15.9% listed any form of
and MV) follow in frequency, and a small number (6.3%) equipment used to measure the radiation absorbed dose,
Fig. 3. Breakdown of the number of authors reporting each physics parameter separated into 4 broad categories. From left
to right, these are source specification, dosimetry/calibration, dose specification, and irradiation geometry.
6 Draeger et al. International Journal of Radiation Oncology Biology Physics
Table 2 Spearman correlation (r), significance (P), and 95% CI between physics-reporting score and various publication metrics
Metric or group All publications Specialist journals* Generalist journals*
Impact factor r Z e0.237 r Z þ0.152 r Z e0.169
P Z 6.9 1024 P Z 1.6 1010 P Z 8.9 1013
CI Z –0.280 to –0.192 CI Z þ0.053 to þ0.247 CI Z –0.220 to –0.117
Total citations r Z e0.102 r Z þ0.020 r Z e0.111
P Z 1.7 105 P Z 0.394 P Z 2.9 106
CI Z –0.148 to –0.055 CI Z e0.079 to þ0.120 CI Z –0.163 to –0.059
Citations per year r Z e0.067 r Z þ0.140 r Z e0.072
P Z 5.2 10-3 P Z 3.8 109 P Z 2.4 103
CI Z –0.113 to –0.020 CI Z þ0.041 to þ0.236 CI Z –0.125 to –0.019
Years since publication r Z e0.094 r Z e0.150 r Z e0.103
P Z 8.4 105 P Z 2.7 1010 P Z 1.5 105
CI Z –0.140 to –0.047 CI Z –0.246 to –0.051 CI Z –0.155 to –0.050
(Co-60, Cs-137 sources) scoring lower than MV and kV x- and preclinical radiation biology research in the past 2
rays and all other sources. A univariate analysis of variance decades. Although 3 previous reviews were performed,21-23
showed a highly significant relationship (P Z 9.8 they were limited in scope. Our results are in broad
10216), with subsequent t tests showing significant dif- agreement with these similar reviews performed by Des-
ferences between all data sets (P .0025) except for kV, rosiers et al,21 Pedersen et al,22 and Stone et al.23 Although
MV, and all remaining sources, for which t tests ranged differences in scoring categories and precise scoring
from P Z .37 to P Z .59. Additionally, the t test between methodology make an exact comparison challenging, the
all remaining sources and Co-60 was only P Z .093. reporting frequencies we found are similar (see Table 3) to
In the course of the review, several publications were these previous reviews.
noted to contain errors in which a quantity was mis- Broadly, the type of radiation and delivered radiation
reported. The most innocuous were simple errors of ter- absorbed dose are nearly universally reported. Approxi-
minology, such as misattributing Co-60 as x-rays instead of mately 60% to 80% report radiation quality/beam energy,
gamma rays or radiation from x-ray generators as gamma model or manufacturer of the irradiator, and irradiation
rays. Although these types of errors do not affect inter- dose rate. Very rarely are details concerning the irradiation
pretability of a study, they introduce ambiguity. Also geometry or irradiator calibration reported. Absolute cali-
common were vague descriptionsdexperiments conducted bration seldom extends beyond specifying the type of
using gamma or x-rays of unknown origin or energy (eg, dosimeter used to measure radiation-absorbed dose, itself
“x-rays delivered from a Siemens x-ray machine”) or solely only reported in one-seventh of publications.
described as “ionizing radiation.” In a few cases, This endemic failure to report basic experimental details
unachievable experiments were described, such as a 150 of the physics and dosimetric irradiation technique of bio-
kVp irradiation performed by a Varian TrueBeam medical logical samples can have further-reaching implications. In
linear accelerator, a system operating at MV energiesd2 all relevant literature, this failure to report or reproduce
orders of magnitude higher than reported. Although the radiation absorbed dose is ascribed to the lack of consul-
TrueBeam is equipped with a kV x-ray imaging source, its tation between the radiation biologists performing the study
maximum energy (140 kVp) and dose rate (w0.1 Gy/ and the qualified radiation physicists able to design,
min38,39) are both lower than that reported (150 kVp, 1.018 document, and validate the irradiation protocol.21,23,32
Gy/min). It is unclear whether the irradiator or beam Furthermore, because plausible but misreported quantities
quality was misreported, but the relative biological effec- are undetectable, reporting of experimental details is no
tiveness and, by implication, the study results are conse- guarantee of their accuracy, and only a comprehensive
quently difficult to reproduce, interpret, or place in context. description of the irradiation protocol can ultimately be
Other papers describe irradiations from nonexistent iso- used to evaluate and reproduce it.
topes beyond a simple typographical error (eg, Cs-62 re- In particular, the very low number of publications
ported vs typical Co-60 or Cs-137). reporting on absolute calibration standards (1.2%) or on
dosimetric measurements (15.9%) implies that few centers
calibrate or verify experimentally the dose delivery of their
Discussion irradiators. If true, this constitutes a potential cause of the
repeated difficulty in achieving standardized accurate and
We performed the first systematic large-scale review of the homogeneous dose delivery in multi-institution dosimetric
current state of physics and dosimetry reporting in basic surveys.22,31-33
8 Draeger et al. International Journal of Radiation Oncology Biology Physics
Cumulative Fraction
1–2.5,n = 347
0.8 0.8
2.5–3.3, n = 344 2–3.5, n = 378
3.3–5, n = 362 3.5–6, n = 334
0.6 0.6
5–50, n = 340 6–400, n = 348
0.4 0.4
0.2 0.2
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Reporting Score Reporting Score
Cumulative Fraction
7–14, n = 344 5–7, n = 351
Cumulative Fraction
0.8 0.8
15–26, n = 342 8–11, n = 312
27–54, n = 349 12–15, n = 358
0.6 0.6
>54, n = 357 16–21, n = 344
0.4 0.4
0.2 0.2
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Reporting Score Reporting Score
kV x-rays, n = 439
Cumulative Fraction
in vivo, n = 736
0.8 0.8 MV x-rays, n = 130
in vitro, n = 961
both, n = 61 All others, n = 211
0.6 0.6 Ukn x/γ-ray, n = 111
No source, n = 132
0.4 0.4
0.2 0.2
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Reporting Score Reporting Score
Fig. 5. Breakdown of the statistical distribution of papers achieving a minimum aggregate physics-reporting score as (a) a
function of impact factor, (b) total number of citations, (c) citations per year, (d) years since publication, (e) journal
specialization and biological model, and (f) source type. In the first 4 cases, the data set was divided into 5 subsets that were
as numerically equal as achievable. In the last 2, categories simply fall as they may. In general, articles with better quan-
titative impact metrics (or less recently published) were statistically more likely to achieve a lower aggregate physics-
reporting score. Journals specializing in radiation oncology, biology, or physics outscored more generalist journals, and
more modern radiation sources (MV or kV x-rays) outsourced traditional sources (Co-60 or Cs-137), but papers that did not
report the source (unknown photons, no source listed) scored drastically lower than all others.
We should stress, however, that although insufficient or protocol and makes no representations on other scientific
inaccurate reporting of the experimental protocol impairs aspects (eg, study design, cell line or animal model selec-
reproducibility, it does not necessarily diminish the quality tion, radiation dose delivered, statistics).
of results or applicability of conclusions. The review is Aside from these findings, the radiation biology and
limited in scope to the reproducibility of the irradiation physics community are acutely aware of the urgency of this
Volume - Number - 2019 20 years of incomplete physics in radiobiology 9
* Not reported.
y
Reported separately for in vitro (left) and in vivo (right) experiments; 104 in vitro and 51 in vivo experiments are involved, though they add to 125
papers total.
z
Reported as “Radio-isotope” and “kV, filtration, HVL” for gamma and kV x-ray sources, respectively.
x
Reported as “Dose details”; unclear if referring to dose only or dose, dose rate, and fractionation.
k
Any detail on source/animal irradiation geometry reported, respectively.
{
Reported as “Field size and shape” and “Geometry of Fields,” respectively.
#
Reported differently under “Absolute dosimetry/calibration” and “Radiation source specification,” respectively.
** Reported differently under “Absolute dosimetry/calibration” and “Determination of dose,” respectively.
yy
Out of an original 3542 evaluated articles.
zz
All issues of Radiation Research between March 2010 and March 2011 (not explicitly enumerated as not all articles may have been analyzed).
problem. In a 2016 report aimed toward improving the the relationship between publication impact metrics and
translation of preclinical research to positive clinical trials, physics-reporting scores, generalist journals continued to
members of the NCI reported that “there is an urgent need exhibit the same trends as before whereas specialist jour-
to improve reproducibility and translatability of pre-clinical nals showed positive trends for impact factor and number of
data” to improve therapeutic agents involving radiation but citations per year and an even more pronounced trend to-
that “one impediment to drug development has been irre- ward higher scores in more recent publications, while
producibility of preclinical data,” which could be improved showing no relationship with total number of citations.
by “fastidious calibration and dosimetry of radiation sour-
ces.”40 Recently, the American Association of Physicists in
Medicine, supported by the Radiation Research Society, Conclusions
established Task Group 319 on the guidelines for accurate
dosimetry in radiobiology experiments.41 Finally, the Na- We performed the first large-scale (>1000 publications)
tional Institute of Allergies and Infectious Diseases and the analysis of the current state and prevalence of physics and
NCI recently formed a program to standardize and monitor dosimetry reporting in radiation biology preclinical and
radiation dosimetry delivery among the entirety of their translational research. Although most peer-reviewed arti-
preclinical research funded by the radiation countermea- cles report basic details on the amount, rate, and type of
sures program. The findings in this study only serve to radiation dose delivered, the majority fail to report other
stress the importance and timeliness of such an initiative. details required for a radiation study to be reproducible,
Finally, a weak but significant negative correlation exists interpretable, and comparable. In general, higher-impact
between the physics-reporting score of an article and its journals and papers with greater numbers of citations were
impact metrics: Well-cited papers published in high- correlated with the reporting of fewer experimental details.
impact-factor journals were most likely to have incom- These findings expose a deep lack of reporting of basic
pletely described methods regarding irradiation protocols. radiation physics concepts in a large proportion of the ra-
This may be because these journals have more stringent diation biology literature. In the context where much of
word limits compared with lower-impact-factor journals. nationally funded preclinical research fails to translate to
Furthermore, specialist journals, in vivo studies, and more clinical applications, and with the increasing role of allied
modern radiation sources such as MV and kV x-rays were fields such as immunology in radiobiology studies, journal
more likely to be well described than generalist journals, editors and reviewers have the responsibility to ensure that
in vitro studies, or traditional sources. This is likely peer-reviewed articles report required experimental details
because these publications are more likely to enlist the help related to physics and dosimetry to ensure the reproduc-
of medical physicists who advise on the design and ibility, interpretability, and comparability of radiation
reporting of irradiation protocols. Indeed, when analyzing research.
10 Draeger et al. International Journal of Radiation Oncology Biology Physics
References 23. Stone HB, Bernhard EJ, Coleman CN, et al. Preclinical data on effi-
cacy of 10 drug-radiation combinations: Evaluations, concerns, and
recommendations. Transl Oncol 2016;9:46-56.
1. Baker M, Penny D. Is there a reproducibility crisis? Nature 2016;533: 24. Dos Santos M, Paget V, Ben Kacem M, et al. Importance of
452-454. dosimetry protocol for cell irradiation on a low X-rays facility and
2. Peng R. The reproducibility crisis in science: A statistical counterat- consequences for the biological response. Int J Radiat Biol 2018;94:
tack. Significance 2015;12:30-32. 597-606.
3. Challenges in irreproducible research. Nature 2015;1-6; special ed. 25. Belley MD, Ashcraft KA, Lee C-T, et al. Microdosimetric and bio-
4. Karp NA. Reproducible preclinical researchdIs embracing variability logical effects of photon irradiation at different energies in bone
the answer? PLoS Biol 2018;16:e2005413. marrow. Radiat Res 2015;184:378-391.
5. Mullard A. Reliability of “new drug target” claims called into ques- 26. Yoshizumi T, Brady SL, Robbins ME, et al. Specific issues in small
tion. Nat Rev Drug Discov 2011;10:643-644. animal dosimetry and irradiator calibration. Int J Radiat Biol 2011;87:
6. Gegley CG, Ellis LM. Raise standards for preclinical cancer research. 1001-1010.
Nature 2012;483:531-533. 27. Kegel P, Riballo E, Kühne M, et al. X-irradiation of cells on glass
7. Liu F-F, Okunieff P, Bernhard EJ, et al. Lessons learned from radiation slides has a dose doubling impact. DNA Repair (Amst) 2007;6:1692-
oncology clinical trials. Clin Cancer Res 2013;19:6089-6100. 1697.
8. Hackam DG, Redelmeier DA. Translation of research evidence from 28. Kirsch DG, Diehn M, Kesarwala AH, et al. The future of radiobiology.
animals to humans. JAMA 2006;296:1731-1732. J Natl Cancer Inst 2018;110:329-340.
9. Ionnidis JP, Kim B, Trounson A. How to design preclinical studies in 29. Gengenbacher N, Singhal M, Augustin HG. Preclinical mouse solid
nanomedicine and cell therapy to maximize the prospects of clinical tumour models: Status quo, challenges and perspectives. Nat Rev
translation. Nat Biomed Eng 2018;2:797-809. Cancer 2017;17:751-765.
10. Hatzis C, Bedard PL, Birkbak NJ, et al. Enhancing reproducibility in 30. Williams JP, Brown SL, Georges GE, et al. Animal models for
cancer drug screening: How do we move forward? Cancer Res 2014; medical countermeasures to radiation exposure. Radiation 2010;173:
74:4016-4023. 557-578.
11. Manolagas SC, Kronenberg HM. Reproducibility of results in pre- 31. Zoetelief J, Broerse JJ, Davies RW, et al. Protocol for x-ray dosimetry
clinical studies: A perspective from the bone field. J Bone Miner Res in radiobiology. Int J Radiat Biol 2001;77:817-835.
2014;29:2131-2140. 32. Seed TM, Xiao S, Manley N, et al. An interlaboratory comparison of
12. Samsa G, Samsa L. A guide to reproducibility in preclinical research. dosimetry for a multi-institutional radiobiological research project:
Acad Med 2018;94:47-52. Observations, problems, solutions and lessons learned. Int J Radiat
13. Freedman LP, Cockburn IM, Simcoe TS. The economics of repro- Biol 2016;92:59-70.
ducibility in preclinical research. PLoS Biol 2015;13:1-9. 33. Zoetelief J, Davies RW, Scarpa G, et al. Protocol for x-ray dosimetry
14. Barton MB, Frommer M, Shafiq J. Role of radiotherapy in cancer and exposure arrangements employed in studies of late somatic effects
control in low-income and middle-income countries. Lancet Oncol in mammals. Int J Radiat Biol 1985;47:81-102.
2006;7:584-595. 34. Moher D, Liberati A, Tezlaff J, et al. Preferred Reporting Items for
15. Barton MB, Jacob S, Shafiq J, et al. Estimating the demand for Systematic Reviews and Meta-Analyses: The PRISMA statement.
radiotherapy from the evidence: A review of changes from 2003 to PLoS One 2009;6:e1000097.
2012. Radiother Oncol 2014;112:140-144. 35. PubMed. Available at: https://www.ncbi.nlm.nih.gov/pubmed/.
16. Miller KD, Siegel RL, Lin CC, et al. Cancer treatment and survivor- Accessed August 30, 2017.
ship statistics, 2016. CA Cancer J Clin 2016;66:271-289. 36. Google Scholar. Available at: https://scholar.google.com/. Accessed
17. Noone A, Howlader N, Krapcho M, et al., eds. SEER cancer statistics November 30, 2018.
review, 1975-2015. Available at: https://seer.cancer.gov/archive/csr/ 37. SCImago. SJR - SCImago Journal & Country Rank [Journal Rank].
1975_2015/. Accessed October 10, 2018. Available at: https://www.scimagojr.com/journalrank.php. Accessed
18. Bryant AK, Banegas MP, Martinez ME, et al. Trends in radiation January 16, 2019.
therapy among cancer survivors in the United States, 2000-2030. 38. Poirier Y, Tambasco M. Experimental validation of a kV source model
Cancer Epidemiol Biomarkers Prev 2017;26:963-970. and dose computation method for CBCT imaging in an anthropo-
19. Smith BD, Haffty BG, Wilson LD, et al. The future of radiation morphic phantom. J Appl Clin Med Phys 2016;17:155-171.
oncology in the United States from 2010 to 2020: Will supply keep 39. Nobah A, Aldelaijan S, Devic S, et al. Radiochromic film based
pace with demand? J Clin Oncol 2010;28:5160-5165. dosimetry of image-guidance procedures on different radiotherapy
20. Steinberg M, McBride WH, Vlashi E, et al. National Institutes of modalities. J Appl Clin Med Phys 2014;15:229-239.
Health funding in Radiation Oncology: A snapshot. Int J Radiat Oncol 40. Coleman CN, Higgins GS, Brown JM, et al. Improving the predictive
Biol Phys 2013;86:234-240. value of preclinical studies in support of radiotherapy clinical trials.
21. Desrosiers M, Dewerd L, Deye J, et al. The importance of dosimetry Clin Cancer Res 2016;22:3138-3147.
standardization in radiobiology. J Res Natl Inst Stand Technol 2013; 41. Kry S, Bazalova-Carter M. Task Group No. 319 - Task Group on the
118:403-418. Guidelines for accurate dosimetry in radiation biology experiments
22. Pedersen KH, Kunugi KA, Hammer CG, et al. Radiation biology (TG319). Available at: https://www.aapm.org/org/structure/?commi
irradiator dose verification survey. Radiat Res 2016;185:163-168. ttee_codeZTG319. Accessed August 11, 2019.