This website is intended for healthcare professionals

Predicting ischaemic stroke risk after TIA: promise and pitfalls

Posted in Stroke Series on 3rd Dec 2014

David Werring, Reader in Clinical Neurology, UCL Institute of Neurology, National Hospital for Neurology and Neurosurgery, Queen Square, WC1N 3BG.

David Werring, Reader in Clinical Neurology,
UCL Institute of Neurology,
National Hospital for Neurology and Neurosurgery,
Queen Square, WC1N 3BG.

Introduction to the ACNR Stroke Series

A key aspect of the revolution in the approach to stroke medicine in recent years has been increased awareness of the very high early risk of ischaemic stroke after TIA, with realisation that a TIA syndrome is thus a unique and golden opportunity to avert future disaster by early investigation and treatment. As part of this revolution, risk scores have been developed and widely enshrined in national guidelines and stroke care pathways, especially the ABCD2 score. Such scores may have a very useful role in patient triage, but are subject to misunderstanding of their intended purpose (e.g they are not diagnostic instruments), and to misuse in clinical practice. In this next article in the Stroke series, Aine Merwick and Peter Kelly give an excellent clear and comprehensive insight into the development and implementation of TIA risk prediction scores, with elegant explanation of the statistical approaches needed, as well as a summary of their limitations and a look to the future of such instruments.

Predicting ischaemic stroke risk after TIA: promise and pitfalls


  • Stroke risk following TIA is highest in the days immediately following a TIA.
  • Risk prediction tools can identify patients at low and high risk of stroke in the short term following TIA.
  • Urgent brain and vessel imaging improves risk prediction following TIA.
  • External validation and cost effectiveness studies of clinical prediction tools such as the ABCD3-I may help demonstrate their utility in everyday clinical practice.

Transient ischaemic attack (TIA) is associated with high risk of early stroke, with stroke rates of 10-13% reported in population studies with routine treatment.1 Early recognition and treatment of TIA provides an ideal opportunity for rapid intervention to prevent stroke and related sequalae. TIA is also an important marker of risk of late stroke recurrence, coronary events and cognitive impairment.2,3,4 When combined with clinical assessment by a trained physician, clinical prediction scores for stroke risk after TIA have the potential to be valuable aids, particularly for identification of patients at highest stroke risk.

In practice identifying which patient may be most high risk is a challenge, and therefore clinical prediction tools can be helpful for answering patient’s questions regarding likelihood of stroke after a TIA.

Clinical prediction tools and ABCD2 score
The ideal characteristics of a predictive score include transportability (also termed generalisability and demonstrated by external validation), good calibration (defined as comparison of observed and predicted event rates for groups of patients) and discrimination (the ability of the risk prediction models to distinguish those who go on to experience an outcome event from those who do not).5,6,7

Epidemiological studies have shown that older age, hypertension, diabetes mellitus, multiple recent TIAs are associated with stroke risk.1,8 Further clinical features including motor weakness, speech disturbance, and symptom duration ≥60minutes were associated with increased stroke risk.1,8,9

Based on the features identified in studies of stroke risk following TIA a simple clinical prediction tool was devised for triage purposes. The ABCD2 clinical prediction score was originally intended for use at the initial evaluation of patients with suspected TIA by primary care and emergency department physicians to aid triage decisions for hospital admission and urgent referral to specialist stroke services.9 The ABCD2 score (age ≥60 years [1 point]; blood pressure ≥140/90mmHg [1 point]; clinical features of weakness [2 points] or speech impairment [1 point]; duration of symptoms ≥60 min [2 points] or 10-59 min [1 point]; diabetes mellitus [1 point]) has been developed based on information obtained on basic clinical examination and history taking.9 The score was deliberately designed not to include information frequently obtained after initial investigations have been performed, as it was designed for use by general practioners and emergency department doctors. The score was designed with the aim of helping to accurately triage patients and specifically to identify which patients may be managed in an outpatient/clinic setting (low risk patients) and identify high risk patients, who may benefit most from hospitalisation and/or prioritised diagnostic investigations and treatments.9 Ideally prediction scores for TIA patients would have high sensitivity and high specificity. To determine the validity of a predication score, its discriminative ability to predict stroke is usually evaluated by receiver-operating characteristic (ROC) analysis and the c-statistic (corresponding to area under the ROC curve) calculated. Ideal discrimination produces a c-statistic of 1.0 whereas discrimination which is no better than chance produces a c-statistic of 0.5.6,7

In the original ABCD2 score derivation study the clinically based score predicted stroke by two days after TIA (c-statistic 0.62-0.79) and by seven days after TIA (c-statistic 0.63-0.83).9 Based on clinical outcome events risk categories were assigned (0-3 low risk, 4-5 moderate risk, 6-7 high risk).9 Current international guidelines for use of the ABCD2 prediction tool, have mostly adopted either a greater than or equal to 4 threshold.10,11

External validity of the ABCD2 score has been demonstrated in a meta-analysis of 11 independent TIA cohorts (ie. excluding the original samples in which the score was derived and validated). On receiver operating characteristic analysis, the pooled area-under-curve (AUC) for seven day stroke was 0.69 (CI 0.64-0.74).12

Carotid stenosis, neurovascular imaging and risk prediction after TIA
Imaging evidence of carotid stenosis (≥50% lumen narrowing) has also been linked with high risk of early recurrence in several studies. In a population based study of 433 TIA patients the hazard ratio (HR) of 90-day stroke recurrence associated with any carotid stenosis >50% was 2.6 (95% CI, 1.28 to 5.20) and with >70% carotid stenosis was 3.3 (95% CI, 1.5 to 7.4, P=0.002).13 The risk of 90-day stroke was seen to rise in a linear fashion with increasing severity of carotid stenosis, ranging from 5.4% (<50% stenosis) to 17.2% (>70% stenosis or occlusion) (P=0.002). Carotid stenosis had moderate sensitivity (43.8%) but high specificity (77.9%) for identification of TIA patients who subsequently developed 90-day stroke. In the OXVASC study patients with posterior circulation TIA, 50% vertebral and basilar stenosis was also associated with 90-day risk of recurrent stroke/TIA (OR 3.2, P=0.006), with rates of 22% for stroke and 46% for TIA or stroke recurrence.14

Figure 1: DWI images showing acutely restricted diffusion in left parietal lobe in a 45 year old patient who presented with a TIA.

Figure 1: DWI images showing acutely restricted diffusion in left parietal lobe in a 45 year old patient who presented with a TIA.

Several groups have demonstrated that addition of brain imaging, especially magnetic resonance imaging (MRI) data may enhance the predictive utility of existing clinical scores.15-18 Diffusion weighted imaging sequences are sensitive to detecting ischaemic changes after TIA (Figure 1). Coutts and colleagues devised an ABCD2 + MRI scoring system in which the ABCD2 score items were retained and two MRI items added (presence of a DWI lesion (1 point) and intracranial vessel occlusion (1 point)). The combined score predicted 90-day recurrent stroke better than ABCD2 (AUC of 0.88 versus 0.78, P=0.01).15 Giles and colleagues conducted a meta-analysis in 4,574 patients imaged with either brain CT or MRI.18 The presence of infarction on DWI was a more powerful predictor of stroke, than ischaemic change on CT. The odds ratio (OR) for stroke with brain infarction on DWI was 14.9 (7.4-30.2) and on CT was 4.2 (2.6-6.9).18

An ideal risk prediction score when compared to an existing score has improved net reclassification (defined as the difference in proportions moving up to a higher score value and down to a lower score value, from one score when compared to another score), and high inter user reliability (low variation in measurements when taken by different users, and high consistency to scoring).7 Furthermore, an ideal clinical prediction score has applicability across different health care settings, clinical credibility and effectiveness (how well a score works in clinical practice) and is straightforward to use in clinical practice.5,6,7

In a pooled analysis of individual patient data from 2,654 TIA patients, a refined clinical and imaging-based prediction score was derived by logistic regression modelling (ABCD3-I) with two points assigned each for ‘dual TIA’ (defined as an earlier TIA within seven days of the TIA prompting medical assessment), positive DWI, and stenosis >50% on carotid imaging.19 The 13-point ABCD3-I score substantially improved predictive ability compared to the ABCD2 score. The ABCD3-I score improved predictive and risk classification of TIA patients with and without stroke (90-day net reclassification improvement 39.4%, p=0.034).19 When the ABCD3-I score was applied in an independent validation sample of 1,232 patients from two population-based studies, the c-statistic increased at each time interval compared to the ABCD2 score (from 0¬∑63 to 0¬∑71 [p=0¬∑045] at seven days; from 0¬∑60 to 0¬∑71, p=0¬∑007 at 90 days).19

The ABCD3-I score has been further independently externally validated in five separate cohorts to date.20-24 In a German single centre hospital based study of 235 patients the score was associated with early in hospital stroke recurrences (p=0.021).20 In a Chinese study of 107 patients the AUC for seven day stroke prediction was 0.74, with a further study among 239 eligible patients, showed an AUC of ABCD3-I scores (0.825; 95% confidence interval, 0.752–0.898) was statistically higher than that of ABCD2 scores (0.694; 95% confidence interval, 0.601–0.786; P<0.001).21,22 In a large multi-centre Spanish study the ABCD3-I score was shown to be a powerful predictor of subsequent stroke with an AUC of 0.83 (95% CI 0.72-0.93) at seven days and 0.69 (95% CI 0.53-0.85) at 90 days.23 Improvement in risk classification by the ABCD3-I score when compared to the ABCD2 score has also been shown in a Japanese population.24

The ABCD3-I score is simple to apply, has been independently externally validated, and has the potential to improve risk stratification after TIA in secondary care settings.

Potential for enhanced risk prediction in TIA patients
Several techniques may have a future role in TIA risk prediction. Lipoprotein-associated phospholipase A2 (LPPLA2), a serum marker of plaque macrophage activation was independently associated with a combined outcome measure of stroke, death, large artery or cardioembolic mechanism in 147 acute TIA patients.25 Other substances have also been suggested as biomarkers for stroke risk prediction but validation and determination of the utility of serum biomarkers remains to be verified.

Combining follow-up imaging after TIA with novel imaging modalities such as perfusion weighted imaging (PWI), or arterial spin label imaging may help better characterise stroke risk.26,27

Stroke prediction in patients with carotid stenosis is an area where risk prediction is particularly important, and remains an ongoing area of clinical relevance. An online calculator of stroke risk based on data from the European Carotid Surgery Trial is available, and is a further adjunct to clinical decision making (

Transcranial Doppler (TCD) may provide prognostic information based on detection of intracranial stenosis, occlusion or micro-embolic signals (MES).28,29 In 1,881 TIA patients followed for one year, increased risk of intracranial revascularisation, stroke, myocardial infarction, or vascular death was associated with intracranial stenosis or occlusion detected on TCD, compared to none (adjusted hazard ratio 2.29).29 Use of 18F fluorodeoxyglucose positron-emission tomography (FDG PET) in large artery stroke may identify high-risk TIA patients based on carotid plaque metabolic activity.30 A study of TIA and minor stroke patients with symptomatic carotid stenosis showed FDG PET uptake predicted early stroke recurrence, independently of stenosis severity.

Long term prediction of stroke risk after TIA is an area that less data is available for, however a Japanese group has recently shown some predictive ability of the ABCD3-I score (c-statistic 0.61) at three years.24 However to predict long-term recurrence after TIA, further research is needed. Further external validation of the ABCD3-I score and determination of the role of prediction tools including the ABCD2 score in decision making regarding model of care eg outpatient versus hospitalisation, and cost effectiveness is an area where further research is needed.31

Pitfalls in risk prediction
Any risk prediction tool is not intended as a substitute for a careful clinical assessment, and may not be applicable for some sub-groups (eg. young patients with non-atherosclerotic TIA, or posterior circulation TIA). A prospective study of 216 consecutive patients with posterior circulation ischaemic stroke or TIA presenting as emergencies, found that using a conventional ABCD2 threshold of ≥4, approximately 30% of patients who had recurrent posterior circulation events within the first 90 days following stroke or TIA were not identified as being high risk.32

To improve risk prediction in posterior circulation TIA further research may focus on external validation of the ABCD2 score in posterior circulation events, or the post-investigation phase ABCD3-I score. Substituting vertebrobasilar stenosis in posterior circulation cases, for the two point scored for carotid stenosis in the ABCD3-I score or incorporating the clinical features vertigo, visual symptoms or ataxia into a risk prediction score may offer potential to refine prediction of stroke following posterior circulation TIA.31,33

A limitation of prognostic tools in general is that the majority of outcome events occur in the low or medium risk groups, since the absolute number of events is greater in the low or medium risk group than in the high risk group ‘the prevention paradox’.34

Most patients with transient brief neurological symptoms are initially assessed by physicians other than stroke specialists. Typically 50% of patients referred by non-specialists to specialist TIA clinics with transient symptoms have confirmed TIA.35,36 Establishing a clear diagnosis of TIA may be difficult and a study that examined inter-rater diagnostic agreement between three fellowship-trained vascular neurologists for 55 TIA patients showed only moderate inter-rater agreement was observed, with an agreement coefficient of 0.46 (0.30 to 0.63) when a two point scale (‘likely’/’unlikely’ TIA) was used.37 This study highlights the subjectivity of TIA diagnosis, even among stroke specialists, and indicates the need for more objective diagnostic measures.37 Using TIA prediction tools in patients that have not actually had a TIA may not be helpful to an individual patient. Transient focal neurological episodes in cerebral amyloid angiopathy, sometimes termed ‘amyloid spells’ can mimic transient ischemic attacks, but are probably more often related to bleeding (especially superficial cortical siderosis or focal convexity sub-arachnoid haemorrhage) rather than ischemia.38 Importantly, such episodes may also herald a high future risk of symptomatic intracerebral lobar haemorrhage, and thus prediction tools for ischaemic stroke would not be anticipated to be useful in this specific scenario.

MRI with diffusion weighted imaging DWI, has been recommended as the preferred imaging modality in TIA, as it has greatest sensitivity for minor ischaemic injury, which informs both diagnosis and risk prediction.19,31 However availability of scan time and clinical contraindications to MRI may limit its use.

Potential pitfall in the use of prediction tools includes delay in presenting with a TIA and inter-rater variability in assigning the score components of the predication tool – e.g. using incorrect value for blood pressure.39

While not intended to replace clinical judgement in the assessment of individual patients, clinically useful risk stratification by prediction tools provide the clinician with an easy to use method of estimating stroke risk.

The ABCD2 clinical prediction tool is a well validated triage tool in TIA management. Risk prediction in the post investigation phase of TIA management benefits from the incorporation of information gained from imaging e.g. presence of imaging abnormalities on parenchymal imaging or detection of large vessel stenosis. The ABCD3-I score which incorporates imaging information has been shown to improved prediction when compared to the ABCD2 score.

Future work may also need to examine the cost effectiveness and safety of using the ABCD2 score predictive tool in clinical decision making in patients with transient neurological symptoms, as well as determine the external validity of the use of the score by non-specialists or non-physicians and via tele-medicine.

Robust external validation of ABCD3-I score, investigation of the safety and use of acute TIA management algorithms based on ABCD3-I, and research on clinical scenarios in which the utility of prediction tools are unclear (e.g. posterior circulation TIA or TIA in young adults) may be useful next steps for refining risk prediction in TIA.

  1. Rothwell PM, Giles MF, Flossmann E, Lovelock CE, Redgrave JN, Warlow CP, et al. A simple score (ABCD) to identify individuals at high early risk of stroke after transient ischaemic attack. Lancet. 2005;366:29–36.
  2. Yang J, Fu JH, Chen XY, Chen YK, Leung TW, Mok V, et al. Validation of the ABCD2 score to identify the patients with high risk of late stroke after a transient ischemic attack or minor ischemic stroke. Stroke. 2010 41(6):1298-300.
  3. Touzé, Varenne O, Chatellier G, Peyrard S, Rothwell PM, Mas JL. Risk of Myocardial Infarction and vascular death after transient ischemic attack and ischemic stroke. A systematic review and meta-analysis. Stroke 2005;36:2748-55.
  4. Takahashi PY, Dyrbye LN, Thomas KG, Cedeno OQ, North F, Stroebel RJ, et al. The association of transient ischemic attack symptoms with memory impairment among elderly participants of the third US National Health and Nutrition Examination Survey. J Geriatr Psychiatry Neurol 2009;22:46-51.
  5. Cook CE. Potential pitfalls of clinical prediction rules. J Man Manip Ther. 2008;16(2):69-71.
  6. Altman DG, Vergouwe Y, Royston P, Moons KGM. Prognosis and prognostic research: validating a prognostic model. BMJ 2009;338:b605
  7. Pencina MJ, D’Agostino RB, Sr., D’Agostino RB, Jr., Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008;27:157-72.
  8. Johnston SC, Gress DR, Browner WS, Sidney S. Short-term prognosis after emergency department diagnosis of TIA. JAMA 2000;284:2901-2906.
  9. Johnston SC, Rothwell PM, Nguyen-Huynh MN, Giles MF, Elkins JS, Bernstein AL, et al. Validation and refinement of scores to predict very early stroke after transient ischaemic attack. Lancet 2007;369:283-292.
  10. National Stroke Foundation. Clinical Guidelines for Acute Stroke Management 2010. Melbourne Australia [Cited 2014 August 21]. Available from
  11. Intercollegiate Stroke Working Party. National clinical guideline for stroke. 4th ed. Royal College of Physicians, 2012. [Cited 2014 August 21]
  12. Giles MF, Rothwell PM. Systematic review and pooled analysis of published and unpublished validations of the ABCD and ABCD2 transient ischemic attack risk scores. Stroke 2010;41:667-73.
  13. Sheehan O, Kyne L, Kelly LA, Hannon N, Marnane M, Merwick A, et al. A population-based study of ABCD2 score, carotid stenosis, and atrial fibrillation for early stroke prediction after TIA. The North Dublin TIA Study. Stroke 2010; 41:844-5.
  14. Marquardt L, Kuker W, Chandratheva A, Geraghty O, Rothwell PM. Incidence and prognosis of > or = 50% symptomatic vertebral or basilar artery stenosis: prospective population-based study. Brain. 2009;132:982-8.
  15. Coutts SB, Eliasziw M, Hill MD, Scott JN, Subramaniam S, Buchan AM et al. An improved scoring system for identifying patients at high early risk of stroke and functional impairment after an acute transient ischemic attack or minor stroke. Int J Stroke 2008;3:3-10.
  16. Ay H, Arsava EM, Johnston SC, Vangel M, Schwamm LH, Furie KL, et al. Clinical- and imaging-based prediction of stroke risk after transient ischemic attack: the CIP model. Stroke 2009; 40:181-6.
  17. Asimos AW, Rosamond WD, Johnson AM, Price MF, Rose KM, Murphy CV, et al. Early Diffusion Weighted MRI as a negative predictor for disabling stroke after ABCD2 score risk categorization in transient ischemic attack patients. Stroke 2009;40:3252-7.
  18. Giles MF, Albers GW, Amarenco P, Arsava MM, Asimos A, Ay H, et al. Addition of brain infarction to the ABCD2 score (ABCD2-I): a collaborative analysis of unpublished data on 4574 patients. Stroke. 2010;41:1907-13.
  19. Merwick A, Albers GW, Amarenco P, et al. Addition of brain and carotid imaging to the ABCD2 score to improve identification of patients at high early stroke risk after transient ischaemic attack. Lancet Neurol. 2010; 9(11):1060-9.
  20. Chatzikonstantinou A, Wolf ME, Schaefer A, Hennerici MG. Risk Prediction of Subsequent Early Stroke in Patients with Transient Ischemic Attacks. Cerebrovasc Dis. 2013;36(2):106-9.
  21. Song XK, Wang WJ, Li HY, Ren MS, Wu L, Ma JF. The value of ABCD3-I score in prediction of cerebral infarction after transient ischaemic attack. Zhonghua Nei Ke Za Zhi. 2012 Jun;51(6):445-8.
  22. Song B, Fang H, Zhao L, Gao Y, Tan S, Lu J, Sun S, Chandra A, Wang R, Xu Y. Validation of the ABCD3-I score to predict stroke risk after transient ischemic attack. Stroke. 2013;44:1244-8.
  23. Purroy F, Jiménez-Caballero PE, Mauri-Capdevila G, Torres MJ, Gorospe A, Ramírez Moreno JM, de la Ossa NP, Cánovas D, Arenillas J, Alvarez-Sabín J, Martínez Sánchez P, Fuentes B, Delgado-Mederos R, Martí-Fàbregas J, Rodríguez Campello A, Masjuán J. PROMAPA study: Stroke Project, Cerebrovascular Diseases Study Group, Spanish Neurological Society.Predictive value of brain and vascular imaging including intracranial vessels in transient ischaemic attack patients: external validation of the ABCD3-I score. Eur J Neurol. 2013;20(7):1088-93.
  24. Kiyohara T, Kamouchi M, Kumai Y, Ninomiya T, Hata J, Yoshimura S, Ago T, Okada Y, Kitazono T. Fukuoka Stroke Registry Investigators. ABCD3 and ABCD3-I scores are superior to ABCD2 score in the prediction of short- and long-term risks of stroke after transient ischemic attack. Stroke. 2014 ;45(2):418-25.
  25. Cucchiara B, Messe S, Sansing L, et al. Lipoprotein-associated phospholipase A2 and C-reactive protein for risk stratification of patients with TIA. Stroke2009;40:2332-2336
  26. Macintosh BJ, Lindsay AC, Kylintireas I, Kuker W, Güther M, Robson MD, et al. Multiple inflow pulsed arterial spin-labeling reveals delays in the arterial arrival time in minor stroke and transient ischemic attack. AJNR Am J Neuroradiol 2010;31:1892-4.
  27. Asdaghi, Negar, et al. Perfusion MR Predicts Outcome in High-Risk Transient Ischemic Attack/Minor Stroke A Derivation–Validation Study. Stroke 2013;44.9:2486-92.
  28. Moustafa RR, Izquierdo-Garcia D, Fryer TD, et al. Carotid Plaque Inflammation Is Associated with Cerebral Microembolism in Patients with Recent TIA or Stroke: A Pilot Study. Circ Cardiovasc Imaging. 2010; (5):536-41.
  29. Meseguer E, Lavallée PC, Mazighi M, et al. Yield of systematic transcranial Doppler in patients with transient ischemic attack. Ann Neurol 2010;68:9-17.
  30. Marnane M, Merwick A, Sheehan OC, Hannon N, Foran P, Grant T, et al. Carotid plaque inflammation on (18) F-fluorodeoxyglucose positron emission tomography predicts early stroke recurrence. Ann Neurol 2012;71:709-18.
  31. Merwick A, Werring DJ. Posterior circulation ischaemic stroke. BMJ 2014;348:g3175.
  32. Gulli G, Markus HS. The use of FAST and ABCD2 scores in posterior circulation, compared with anterior circulation, stroke and transient ischemic attack. J Neurol Neurosurg Psychiatry. 2012;83:2 228-229.
  33. Gulli G, Marquardt L, Rothwell PM, Markus HS. Stroke risk after posterior circulation stroke/transient ischemic attack and its relationship to site of vertebrobasilar stenosis: Pooled data analysis from prospective studies. Stroke. 2013;44:598-604.
  34. Stewart LA, Clarke MJ. Practical methodology of meta-analyses (overviews) using updated individual patient data. Cochrane Working Group. Stat Med 1995;14:2057–79.
  35. Quinn TJ, Cameron A, Dawson J, Lees KR, Walters MR. ABCD2 scores and prediction of noncerebrovascular diagnoses in an outpatient population. A case-control study. Stroke 2009;40:749-53.
  36. Sheehan OC, Merwick A, Kelly LA, Hannon N, Marnane M, Kyne L. et al. Diagnostic usefulness of ABCD2 score to distinguish TIA and minor ischemic stroke from non-cerebrovascular events: the North Dublin TIA Study. Stroke. 2009;40:3449-54.
  37. Castle J, Mlynash M, Lee K, et al. Agreement regarding diagnosis of transient ischemic attack fairly low among stroke-trained neurologists. Stroke. 2010;41:1367-70.
  38. Charidimou A, Gang Q, Werring DJ. Sporadic cerebral amyloid angiopathy revisited: recent insights into pathophysiology and clinical spectrum. J Neurol Neurosurg Psychiatry 2012;83:124-13.
  39. Bradley D, Cronin S, Kinsella JA, Tobin WO, Mahon C, O’Brien M, Lonergan R, Cooney MT, Kennelly S, Collins DR, O’Neill D, Coughlan T, Smyth S, McCabe DJ. Frequent inaccuracies in ABCD2 scoring in non-stroke specialists’ referrals to a daily Rapid Access Stroke Prevention service. J Neurol Sci. 2013;332(1-2):30-4.

ACNR 2015;14(6):8-11. Published online first: 3 December, 2014

Download this Article