Objective Differentiation between progressive multifocal leukoencephalopathy (PML) and new multiple sclerosis (MS) lesions on brain MRI during natalizumab pharmacovigilance in the absence of clinical signs and symptoms is challenging but is of substantial clinical relevance. We aim to define MRI characteristics that can aid in this differentiation.
Methods Reference and follow-up brain MRIs of natalizumab-treated patients with MS with asymptomatic PML (n=21), or asymptomatic new MS lesions (n=20) were evaluated with respect to characteristics of newly detected lesions by four blinded raters. We tested the association with PML for each characteristic and constructed a multivariable prediction model which we analysed using a receiver operating characteristic (ROC) curve.
Results Presence of punctate T2 lesions, cortical grey matter involvement, juxtacortical white matter involvement, ill-defined and mixed lesion borders towards both grey and white matter, lesion size of >3 cm, and contrast enhancement were all associated with PML. Focal lesion appearance and periventricular localisation were associated with new MS lesions. In the multivariable model, punctate T2 lesions and cortical grey matter involvement predict for PML, while focal lesion appearance and periventricular localisation predict for new MS lesions (area under the curve: 0.988, 95% CI 0.977 to 1.0, sensitivity: 100%, specificity: 80.6%).
Interpretation The MRI characteristics of asymptomatic natalizumab-associated PML lesions proved to differ from new MS lesions. This led to a prediction model with a high discriminating power. Careful assessment of the presence of punctate T2 lesions, cortical grey matter involvement, focal lesion appearance and periventricular localisation allows for an early diagnosis of PML.
Statistics from Altmetric.com
Pharmacovigilance of patients with multiple sclerosis (MS) includes the detection of disease activity, comorbidities (eg, vascular pathology) and drug-related adverse events including opportunistic infections. For these purposes, brain MRI plays an essential role, in addition to monitoring of clinical symptoms and paraclinical tests (eg, cerebrospinal fluid (CSF) testing and blood biomarkers).1–3 The introduction of natalizumab, a humanised monoclonal antibody against the α4-integrin adhesion molecule approved for the treatment of relapsing MS, has further stressed the need for strict pharmacovigilance, including brain MRI monitoring.4–6 Although clinically highly effective, the use of natalizumab is associated with an increased risk for JC virus (JCV)-related diseases such as progressive multifocal leukoencephalopathy (PML), an opportunistic infection of glial and neuronal cells in the central nervous system caused by reactivation and replication of the JCV and granule cell neuronopathy.4 ,7–13
Early detection of natalizumab-associated PML is clinically relevant because patients with PML who are asymptomatic at the time of diagnosis have a better functional outcome and survival compared to those who are symptomatic.14–16 Several reports have indeed demonstrated that brain MRI may detect natalizumab-associated PML lesions several months before the patient develops symptoms (asymptomatic natalizumab-associated PML lesions).14–22 Therefore, the use of brain MRI in pharmacovigilance of natalizumab treated patients with MS for screening purposes with respect to the early detection of PML in high-risk patients has been recommended by recent expert panel guidelines.23–25
When a new brain lesion is detected on MRI in clinically stable patients with MS during natalizumab treatment, the crucial but also challenging question is whether this is a new asymptomatic PML lesion or a new MS lesion. Since JCV DNA is frequently undetectable in the CSF of these asymptomatic patients with PML, it is of paramount importance to accurately make this distinction based on MRI findings.20 On brain MRI, new asymptomatic PML lesions may be rather small and mimic other pathologies, especially MS lesions, as they can share common characteristics.20 ,26–28 This allows for asymptomatic PML lesions to be missed or misinterpreted as MS lesions, with clinical consequences.
The challenge of diagnosing asymptomatic PML and differentiating asymptomatic PML lesions from new MS pathology has been stressed by a previous study showing that the sensitivity for the detection of asymptomatic PML was only 59.5% with a specificity of 91.7%.28 In addition, there was moderate agreement among the readers with respect to the detection of asymptomatic PML (Fleiss' κ of 0.52).28 ,29 The two main factors contributing to the low sensitivity appeared to be the inability to detect smaller new asymptomatic PML lesions and the misinterpretation of asymptomatic PML lesions as new focal MS lesions. Thus, the identification of discriminative MRI lesion characteristics that differentiate asymptomatic PML from new MS lesions during pharmacovigilance is of utmost clinical importance. The aim of this study was to identify these MRI lesion characteristics.
Patients and methods
Standard protocol approvals, registrations and patient consents
Brain imaging of patients with MS treated with natalizumab is part of the standard patient care. A waiver was provided by the local institutional review board (IRB), stating that the requirements of the Medical Research Involving Human Subjects Act (WMO) did not apply to this study and that an official IRB approval was not required. We obtained written informed consent from all included patients for the use of the clinical, laboratory and MRI data for research and educational purposes.
Study design and patient selection
In this retrospective follow-up study, we further analysed data obtained following a previous study, which included asymptomatic natalizumab-associated patients with PML with MS, and patients with MS either with or without new MS lesions on follow-up brain MRI during natalizumab treatment.28 The study design, patient selection and patient characteristics of the study groups have been described in the previous study.28 In short, the asymptomatic PML group comprised patients from the Dutch-Belgian natalizumab-associated PML database as well as patients who were referred to us for a second opinion and research purposes. This group included 21 patients, 15 of whom were female. Mean age: 45 years (SD: 11 years); mean natalizumab treatment duration: 44 months (SD: 18 months); median of two new lesions per patient (range 1–5). These patients showed no new concomitant MS lesions on the follow-up MRI. The patients with new MS lesions during natalizumab treatment were selected from our local cohort of natalizumab-treated patients with MS. This group included 20 patients, 14 of whom were female. Mean age: 38 years (SD: 8 years); mean natalizumab treatment duration: 11 months (SD: 4 months); median of two new lesions per patient (range 1–6). These patients showed no new concomitant PML lesions on the follow-up MRI. Given the aim of the current study, patients without new lesions on follow-up MRI were not included.
MRI protocols and readers
The detailed information regarding MRI protocols (including at least axial T2/proton density-weighted or fluid-attenuated inversion recovery (FLAIR) images in all patients) used in this study and the clinical experience of the blinded readers who performed the ratings have been described in the previous study.28 Diffusion-weighted images (DWI) were not included in this study because, although DWI were mostly present in the asymptomatic PML group, these images were not available in the new MS lesion group.
The image analysis for this study was performed in a second evaluation subsequent to the first image analysis step (to determine for each patient if new lesions were present on the follow-up scan compared to the reference scan, and whether the readers considered the lesions to be suggestive for PML or MS pathology), which was analysed in the previous study.28 In the second evaluation, each reader had to score the lesion characteristics of all newly detected lesions according to a specially designed scoring scheme based on previously reported PML and MS lesion characteristics.21 ,22 ,26 ,30 ,31 The proposed characteristics were: the lesion distribution (unilobar (lesions confined to one lobe), multilobar (lesions in two or more contiguous lobes) or widespread (lesions in two or more non-contiguous lobes or present in both hemispheres)); hemispherical involvement (unilateral or bilateral); anatomical location (frontal lobe, parietal lobe, occipital lobe, temporal lobe, corpus callosum, thalamus, basal ganglia, brainstem or cerebellum), tissue involvement (juxtacortical white matter, deep white matter, periventricular white matter, cortical grey matter or deep grey matter), lesion appearance (focal, diffuse, confluent irregular or infiltrative), lesion borders towards the white and grey matter (sharp, mixed or ill-defined), lesion size (>3 cm or <3 cm in diameter), signs of mass effect (present or absent), signs of oedema (present or absent), contrast enhancement (present or absent), intensity on T1, T2 and FLAIR images (hypointense, isointense, hyperintense), and the presence of punctate T2 lesions (punctiform T2/FLAIR hyperintense lesions in the vicinity of the main PML lesions recently described as ‘milky way appearance’ and/or punctiform T2/FLAIR hyperintense lesions with a perivascular spread, <5 mm in diameter).
Using generalised estimating equations (GEEs) in SPSS statistics software V.22 (IBM Corp, Armonk, New York, USA), we investigated the association of the lesion characteristics with asymptomatic PML or new MS lesions. A logit function was used to link the characteristics of the lesion nested within a patient to the outcome (asymptomatic PML group or new MS lesion group), and an independent correlation structure was chosen as two separate lesions of the same cause within a patient will probably have different lesion characteristics. Since the ratings of the current and previous study were performed consecutively, we were unable to include those lesions that were missed by the readers in the first image analysis step (as described in the previous study)28 in the current analyses. Therefore, the number of rated patients in both groups is not the same for each of the four raters. For each separate lesion characteristic, the association with asymptomatic PML or new MS lesions was assessed. Results were expressed as an OR with accompanying 95% CI, and p value for the test of model effects. An OR >1 indicates an association with asymptomatic PML; conversely, an OR ≤1 indicates an association with new MS lesions. For the categorical variables, lesion borders towards the white and grey matter, and the lesion dissemination, the categories sharp borders and a widespread lesion dissemination were used as reference category, respectively. Only overall p values were reported for these variables.
For those lesion characteristics with complete data separation (ie, in either the PML or the MS group, the characteristic is either present or absent for all cases), we tested the association using the Pearson χ2 or Fisher's exact test. We thereby ignored the repeated scorings within a patient and were unable to calculate an OR.
We built a multivariable prediction model consisting of lesion characteristics predictive for asymptomatic PML or new MS lesions via a forward selection procedure (p value for entry <0.05). The model was corrected for the potential confounding patient characteristics age, gender and treatment duration. Using the Receiver Operating Characteristic (ROC) curve, the ability of the model to discriminate between asymptomatic PML and new MS lesions was assessed. The ROC curve was based on all patient ratings of each separate rater combined. When multiple lesions were rated within a patient, the lesion with the highest value of the linear predictor was used for analysis, which is the lesion that is most predictive for PML and would most likely guide treatment decision-making.
Finally, we calculated a Cohen's κ between all sets of raters for the lesion characteristics that were significantly associated with asymptomatic PML or new MS lesions in order to test the inter-rater reliability for the scoring of these lesion characteristics. For the categorical characteristics, we calculated Fleiss' κ between the sets of raters. The κ values were interpreted according to the Landis and Koch scheme: κ<0.00, no agreement; 0.00≤κ≤0.20, slight agreement; 0.21≤κ≤0.40, fair agreement; 0.41≤κ≤0.60, moderate agreement; 0.61≤κ≤0.80, substantial agreement; 0.81≤κ≤1.00, almost perfect agreement.29 ,32
For all tests, a p value of <0.05 was considered statistically significant.
Association of the separate lesion characteristics with asymptomatic PML or new MS lesions
We calculated the rating frequencies of all lesion characteristics, the ORs for the association of these lesion characteristics with asymptomatic PML, and the accompanying 95% CIs and p values (table 1).
Presence of punctate T2 lesions showed a very strong association with asymptomatic PML lesions (OR: 135.6, CI 18.9 to 974.7, p<0.001, figure 1). New lesions with cortical grey matter or juxtacortical white matter involvement were more likely to be asymptomatic PML lesions than MS lesions (OR: 14.8, CI 3.6 to 61.1, p<0.001, and OR: 3.6, CI 1.2 to 11.1, p=0.023, respectively, figure 2). Both ill-defined and mixed borders (compared to the reference category sharp borders) towards the white matter, and towards the grey matter, were significantly associated with asymptomatic PML (OR: 9.1, CI 4.0 to 20.5, and OR: 6.2, CI 1.9 to 20.5, p<0.001, respectively, for white matter and OR: 3.7, CI 1.6 to 8.8, and OR: 4.8, CI 1.6 to 14.5, p=0.001, respectively, for grey matter). Furthermore, presence of contrast enhancement was significantly associated with asymptomatic PML (OR: 9.5, CI 1.1 to 82.2, p=0.041, figure 2).
Conversely, new lesions with a focal appearance (as opposed to a diffuse, confluent irregular or infiltrative appearance), and lesions located in the periventricular white matter were more likely to be new MS lesions instead of asymptomatic PML (OR: 0.026, CI 0.0081 to 0.083, p<0.001, and OR: 0.028, CI 0.0036 to 0.22, p=0.001, respectively, figure 3).
Multivariable prediction model of lesion characteristics differentiating asymptomatic PML from new MS lesions
A multivariable model of lesion characteristics predicting a new lesion, being either asymptomatic PML or a new MS lesion, was constructed (table 2). The model was corrected for age and treatment duration.
Presence of punctate T2 lesions and cortical grey matter tissue involvement remained significant predictors for asymptomatic PML lesions in the multivariable model (OR:183.2, CI 11.4 to 2950.7, p<0.001 and OR: 59.8, CI 8.4 to 427.6, p<0.001). A focal lesion appearance and periventricular white matter tissue involvement remained significant predictors for new MS lesions in the multivariable prediction model (OR: 0.009, CI 0.0008 to 0.12, p<0.001, and OR: 0.0006, CI 0.00003 to 0.0121, p<0.001, respectively). The algebraic expression of the model is available as online supplementary material.
We calculated an ROC curve for the multivariable model to test the ability of the model to discriminate between asymptomatic PML and new MS lesions (figure 4). The model showed a very high discriminating power between asymptomatic PML and new MS lesions with an area under the curve (AUC) of 0.988 (95% CI 0.977 to 1.0), based on 65 ratings of asymptomatic patients with PML and 62 ratings of patients with new MS lesions by the four raters combined. Using a linear predictor threshold of −2.15 corresponds to a sensitivity of 100% with a specificity of 80.6%. Likewise, applying a linear predictor cut-off of 3.04 corresponds to a sensitivity of 86.2% with a specificity of 100%. The ROC curves for each rater separately are available as online supplementary material.
Inter-rater agreement on lesion characteristics
The inter-rater agreement on the scoring of all significant lesion characteristics was calculated (table 3). The agreement is presented as the median and the range of Cohen's κ between all sets of two raters.
Of the lesion characteristics included in the multivariable model, a focal lesion appearance, periventricular white matter localisation and punctate T2 lesions showed a moderate inter-rater agreement (median κ: 0.56, 0.59 and 0.57, respectively), and cortical grey matter showed a fair inter-rater agreement (median κ: 0.22).
Among the other lesion characteristics that were associated with asymptomatic PML, juxtacortical white matter involvement and presence of contrast enhancement showed a substantial inter-rater agreement, lesion size showed a moderate inter-rater agreement and lesion borders towards white and grey matter showed just a slight inter-rater agreement.
This study showed that in the absence of clinical signs and symptoms it is possible to differentiate PML lesions from new MS lesions based on a careful assessment of lesion characteristics on MRI in patients with MS treated with natalizumab. A recent study by Hodel et al33 comparing natalizumab-treated patients with MS without new lesions to patients with asymptomatic natalizumab-associated PML described useful MRI features predictive for asymptomatic PML lesions (DWI hyperintensity and U fibre involvement). In our study, we tried to make one important step further by simulating the real-life pharmacovigilance setting with reference MRI scans and aimed to directly compare the lesion characteristics of new asymptomatic PML lesions and new MS lesions in natalizumab-treated patients with MS. We describe rather conclusive MRI lesion characteristics which can aid in differentiating asymptomatic PML lesions from new MS lesions in a pharmacovigilance setting of natalizumab-treated patients with MS, thereby potentially increasing the chance of an early and specific diagnosis of asymptomatic PML based on brain MRI.
The presence of punctate T2 lesions showed a very strong association with asymptomatic PML lesions. In previous studies, punctiform T2/FLAIR hyperintense lesions in the vicinity of the main PML lesions, also described as ‘milky way appearance’ when being present in high numbers,22 showed to be common in symptomatic and asymptomatic natalizumab-associated PML.21 ,22 ,30 ,33 Punctiform T2 lesions are also seen further away from the main PML lesion, even in the hemisphere contralateral to the main PML lesion, and can have a perivascular distribution pattern suggestive of perivascular inflammation.22 ,26 ,30 In clinical practice, it can be difficult to differentiate between punctiform T2 lesions in the vicinity of the main PML lesion and punctiform T2 lesions with a perivascular spread. In addition, it is likely that there is an overlap in these two types of punctate lesions and they may share similar characteristics in terms of the underlying pathology.34 ,35 Therefore, we did not differentiate between these two types of punctate T2 lesions and considered the presence of any punctate T2 lesions as one characteristic. In line with these results, a recent study showed that these punctate T2 lesions are frequently visible in various stages of natalizumab-associated PML as well as in PML due to other conditions, while clinically isolated syndrome and patients with MS do not show these punctate T2 lesions.36 Given the compiling evidence from our study in addition to the previous studies, the presence of punctate T2 lesions seems to be one of the most important imaging features when differentiating asymptomatic PML lesions from new MS lesions. The presence of these punctate T2 lesions may lead to a specific diagnosis of asymptomatic PML.
Even though MS lesions are known to occur in the juxtacortical white and cortical grey matter (juxtacortical lesions have even been incorporated in the revised MS diagnostic criteria),31 cortical grey matter and juxtacortical white matter lesion involvement were significantly associated with asymptomatic PML instead of new MS lesions in our study. Indeed, the recent study by Hodel et al33 nicely showed that a subcortical lesion location, involving the U fibres, was predictive for asymptomatic natalizumab-associated PML. In contrast, a focal lesion appearance and lesions located in the periventricular white matter were strongly associated with new MS lesions instead of asymptomatic PML, which is not surprising since this type of lesion is one of the well-known stereotype MS lesions included in the MRI diagnostic criteria for MS.31 In our study, we found no significant difference in lesion intensity on T1, T2 or FLAIR images between asymptomatic PML and new MS lesions. However, the presence of contrast enhancement was significantly associated with asymptomatic PML. This might be due to the fact that the strong treatment effect of natalizumab prevents blood-brain barrier breakdown, and therefore new MS lesions during natalizumab treatment are less likely to show any contrast enhancement. Thus, contrast-enhancing lesions during natalizumab treatment are more likely inflammatory PML lesions, a phenomenon seen in ∼30% of the patients at PML diagnosis.17 ,21 ,22 ,30 ,37
In previous reports, PML lesions have been described as having ill-defined borders towards the white matter and sharp borders towards the grey matter.22 Interestingly, in our data, ill-defined and mixed borders compared to sharp borders towards both the white and the grey matter were significantly associated with asymptomatic PML. This might be due to the fact that the previously reported observation of sharp borders towards the grey matter is mainly based on research in symptomatic patients with PML, who might show more extensive demyelination in the juxtacortical white matter compared to asymptomatic patients with PML, thereby creating a more visible contrast between the grey matter and the lesion.22
A lesion size of >3 cm was frequently rated in the asymptomatic PML lesions and in none of the new MS lesions. Owing to the complete data separation in this lesion characteristics, we tested the association via the Pearson χ2 test, which showed a significant association (p<0.001). Although we thereby erroneously ignore the repeated scorings within a patient, and are unable to calculate an OR, it does indicate that this lesion characteristic is highly relevant. In previous reports, it has also been demonstrated that symptomatic PML lesions are often larger than 3 cm in diameter as opposed to most focal MS lesions.22 ,38 Although Fisher's exact test gives a significant association of the thalamic and corpus callosal lesion localisation with asymptomatic PML and new MS lesions, respectively, the small amount of ratings of the presence of these lesion characteristics calls into question the true meaning of these results. In fact, this might very well be the result of the fact that these tests disregard the multiple ratings within each patient and consider each rating as a single patient.
Ultimately, four lesion characteristics made the p<0.05 threshold in the forward selection procedure of the multivariable prediction model, which was corrected for age and treatment duration (table 2). The presence of a focal lesion appearance and periventricular localisation remained significant in this multivariable model and predict for new MS lesions. Punctate T2 lesions and cortical grey matter involvement significantly predict for asymptomatic PML in our multivariable model. When tested in this data set using an ROC curve, the model showed a very high discrimination power with an AUC of 0.988 and a sensitivity and specificity of 100% and 80.6%, respectively, when the threshold is chosen for maximum sensitivity (figure 4).
The inter-rater agreement for the scoring of the lesion characteristics in the multivariable model was moderate for focal lesion appearance, periventricular white matter localisation and punctate T2 lesions. However, the raters showed a markedly lower (fair) agreement on the scoring of cortical grey matter involvement, indicating that this lesion characteristic is more difficult to score, and thus there might be a need for dedicated training on the lesion characteristics (for instance, via e-learning programmes such as https://ms-pml.org/).
A limitation of our study is that we had to use GEEs to correct for multiple ratings within a patient. Unfortunately, this type of analysis precludes the testing of characteristics with complete data separation (such as lesion size of >3 cm), which were therefore not eligible for the multivariable model. Another limitation is that since the cases were collected retrospectively from a multicentre database, not all MRI scans were performed according to a standardised MRI protocol. However, this is hard to avoid as currently there are no MRI data of a substantial number of asymptomatic patients with PML from a single centre are lacking. In addition, we cannot exclude that when a new lesion was suspected to be either PML or MS, the lesion characteristics were scored in accordance with the suspected cause of the lesion. This may have led to an overestimation of the discriminatory power of the lesion characteristics. The ROC curve of the model was based on the same data as that on which the model was designed, therefore potentially leading to overfitting of the data. In order to validate the model, it should be tested in an independent data set. We were unable to include DWI in this study, which is unfortunate as a subset of asymptomatic PML lesions can be visible on high B-value DWI without showing a low signal on the apparent diffusion coefficient, meaning no real diffusion restriction is present.21 Although new MS lesions are also frequently hyperintense,39 there might still be an added value of DWI in differentiating asymptomatic PML from new MS lesions in a pharmacovigilance setting. However, this needs to be further investigated.
In conclusion, this study identifies MRI lesion characteristics which aid in differentiating asymptomatic PML lesions from new MS lesions in natalizumab-treated patients with MS who do not show new neurological symptoms. Most importantly, we present a prediction model of four lesion characteristics that shows a very high discriminating power in making this distinction. In our model, the presence of punctate T2 lesions and cortical grey matter involvement strongly predict for the lesion being asymptomatic PML, while the presence of a focal lesion appearance and a periventricular location strongly predict for the lesion to be a new MS lesion. Knowledge on these lesion characteristics can improve an early and specific diagnosis of natalizumab-associated asymptomatic PML and thereby lead to a better chance of survival and an improved functional outcome in these patients.
The authors wish to thank all patients included in the study for agreeing to the use of their MRI scans and paraclinical data for research and education purposes. In addition, they would like to thank the following physicians for sharing clinical, imaging and paraclinical data on asymptomatic patients with PML included in this study: Bob W van Oosten (VU University Medical Center, Amsterdam, the Netherlands), Dorine A Siepman (Erasmus MC, University Medical Center Rotterdam, the Netherlands), Jop Mostert (Rijnstate Hospital, Department of Neurology, Arnhem, the Netherlands), Wibe Moll (Maasstad Hospital, Rotterdam, the Netherlands), Alex EL van Golde (ZGT Hospital, Almelo, the Netherlands), Stephan TFM Frequin (St Antonius Hospital, Nieuwegein, the Netherlands), Paul AD Bouma (Tergooi, Blaricum, Hilversum, the Netherlands), and Bénédicte Quivron (CH Jolimont, La Louvière, Belgium). They would like to thank Tineke van IJken, Ronald van Schijndel, Tabe Kooistra and Marlieke de Vos for their administrative and technical support. This research has been executed within the VUmc MS Center Amsterdam.
Contributors MTW and MPW collected and analysed the data and wrote the manuscript. BIW performed the statistical analysis and edited the manuscript. AV collected and interpreted the data and edited the manuscript. SDR, ES, YL and COMJ performed the imaging analysis, interpreted the data and edited the manuscript. FB, BMJU and JK interpreted the data and edited the manuscript. All authors reviewed the manuscript.
Funding The MS Centre Amsterdam is funded by a programme grant (14-358e) from the Stichting voor MS Research (Voorschoten, The Netherlands). YL was supported by the ECTRIMS-MAGNMIS Fellowship from ECTRIMS, the Beijing Nova Programme (xx2013045), the National Science Foundation of China (number 81101038) and Key Projects in the National Science & Technology Pillar Program during the Twelfth Five-Year Plan Period ( number 2012BAI10B04). These funders had no involvement in the conduct of this study.
Competing interests MPW has received consultancy fees from Biogen and Roche. FB serves as a consultant for Bayer-Schering Pharma, Sanofi-Aventis, Biogen, Teva, Novartis, Roche, Synthon BV, Genzyme and Jansen Research. JK has accepted consulting fees from Merck-Serono, TEVA, Biogen, Genzyme and Novartis. BMJU has received consultancy fees from Biogen Idec, Genzyme, Merck Serono, Novartis, Roche and TEVA. AV has received speaking fees from Teva. MTW, ES, COM, YL, SDR and BIW do not report any competing interest. The VUmc MS Center Amsterdam has received financial support for research activities from Bayer Schering Pharma, Biogen, Glaxo Smith Kline, Merck Serono, Novartis and Teva.
Ethics approval Medical Ethics Commitee (METC), VU Medical Center Amsterdam.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.