Article Text
PDF
Abstract
Background In stroke due to middle cerebral artery (MCA) occlusion, collaterals may sustain tissue in the peripheral MCA territory, extending the time window for recanalizing therapies. However, MCA occlusions often block some or all of the ‘lenticulostriate’ (LS) arteries originating from the M1 segment, eliminating blood flow to dependent territories in the striatum, which have no collateral supply. This study examines whether mechanical thrombectomy (MTE) can avert imminent striatal infarction in patients with acute MCA occlusion.
Methods 279 patients with isolated MCA occlusion subjected to MTE were included. Actual LS occlusions and infarctions were assigned to predefined ‘LS occlusion’ and ‘LS infarct’ patterns derived from known LS vascular anatomy. The predictive performance of LS occlusion patterns regarding ensuing infarction in striatal subterritories was assessed by standard statistical measures.
Results LS occlusion patterns predicted infarction in associated striatal subterritories with a positive predictive value (PPV) of 91% and a negative predictive value of 81%. In 15 of the 22 patients who did not develop the predicted striatal infarctions, reassessment of angiographies revealed LS vascular supply variants that explained these ‘false positive’ LS occlusion patterns, raising the PPV to 96%. Symptom onset to recanalization times were relatively short, but this alone could not account for the false positive LS occlusion patterns in the remaining seven of these patients.
Conclusions With currently achievable symptom onset to recanalization times, striatal infarctions are determined by MCA occlusion sites and individual vascular anatomy, and cannot normally be averted by MTE, but there are exceptions. Further study of such exceptional cases may yield important insights into the determinants of infarct growth in the hyperacute phase of infarct evolution.
- Stroke
- Thrombectomy
Statistics from Altmetric.com
Introduction
Collaterals can sustain brain tissue distal to the occlusion site in ischemic stroke caused by large vessel occlusion.1–3 Collateral blood flow modulates the time dependency of tissue damage and may extend the temporal window in which recanalizing therapies are clinically effective.4 Accordingly, the delay between stroke onset and recanalization by mechanical thrombectomy (MTE) has only limited impact on clinical outcome in patients with strong collaterals.5
However, not all brain areas are equally accessible to collateral supply. The basal ganglia, specifically the striatum, and adjacent white matter, usually receive their blood supply from ‘lenticulostriate’ arteries (LSAs) arising from the M1 segment of the middle (MCA), and—directly and via the recurrent artery of Heubner (rHA)—the anterior cerebral artery (ACA). 6 ,7 These LSAs are ‘endarteries’ that do not anastomose with other perforators or peripheral arteries.7–9 If they are blocked by thrombus within their parent vessel segment, blood flow to the dependent tissue will be shut off completely, even if the thrombus does not extend into the LSAs themselves.
The usual vascular supply zones of the LSAs are arranged in an orderly manner.6 ,7 ,10–13 The most medial LSAs from the ACA and the medial M1 segment supply the mediobasal and medial part, respectively, of the caudate head. The more lateral LSAs from the middle M1 segment supply the posterolateral caudate head, the anterior caudate body and tail, and the anterior putamen, while the most lateral LSAs supply the posterior caudate tail and the posterior putamen. The medial LSAs also supply the anterior limb of the internal capsule (IC), and the lateral LSAs may contribute blood supply to the external part of the globus pallidus.7 ,14 The main blood supply of the globus pallidus, however, and the blood supply of the lower part of the genu and posterior limb of the IC, posteromedial to the lentiform nucleus, usually arise from sources other than the LSAs— that is, the anterior choroidal artery, the rHA, and perforators from the posterior communicating artery or the posterior cerebral artery P1 segment.15–17
Due to the lack of external collaterals, this orderly vascular anatomy predisposes to distinct LS infarction patterns depending on the precise site of occlusion.8 ,12 Here we examine, in a large cohort of stroke patients with MCA occlusion, treated with MTE in a routine clinical setting, if, and under which circumstances, the expected infarctions do or do not occur. We hypothesized that they could rarely, if ever, be prevented, even with successful recanalization.
Patients and methods
We retrospectively assessed all patients with ischemic stroke subjected to DSA for MTE in our institution between January 2007 and July 2015. All consecutive patients with isolated MCA occlusion, as determined in preinterventional CT angiography or MR angiography and/or by the initial angiographic (DSA) runs prior to MTE, were included, provided that adequate postinterventional control images were available: either non-enhanced CT (NECT) at least 18 h after symptom onset or MRI (usually within 3–5 days). All procedures were approved by the local ethics committee.
Endovascular therapy
According to institutional guidelines, patients with acute large vessel occlusion stroke were eligible for MTE if there were no extensive early ischemic signs in preinterventional imaging (greater than one third of the dependent territory), and the intervention could begin within 6 h after symptom onset, or within 3 h after recognition of symptoms in patients with an unclear onset time. No age limit was applied. Intravenous thrombolysis with recombinant tissue plasminogen activator (IV rtPA) was applied as bridging therapy in all eligible patients. MTE was performed with standard techniques, as described previously.18 Procedures involved stent retrievers, direct aspiration techniques with large lumen aspiration catheters (ADAPT19), which were used as a standalone technique or in conjunction with stent retrievers, and rarely ‘older’ devices such as the MERCI retriever. Treatments with stent retrievers or direct aspiration/ADAPT were considered as procedures with ‘up to date’ equipment. Successful recanalization was defined as Thrombolysis in Cerebral Infarction (TICI) 2b or 3 on the TICI scale,20 with TICI 2b defined as reperfusion in more than two-thirds of the initially involved territory.
Classification of vascular occlusion and infarct patterns
Following the present concepts of LS vascular anatomy, we predefined four LS vascular occlusion patterns, LS occlusion I–IV, depending on the beginning of the occlusion and the corresponding involvement of the MCA LSAs (LS occlusion I=involvement of the proximal, middle, and distal LSAs; LS occlusion II=middle and distal LSAs; and LS occlusion III=distal LSAs) (figure 1). LS occlusion IV denoted distal MCA occlusions not involving any LSAs. Three LS subterritories T1–T3 and four associated LS infarction patterns, LS infarct I–IV were analogously predefined according to the normal vascular supply zones of the medial, middle, and lateral MCA LSAs. Subterritory T1 comprises the caudate head (excluding the most mediobasal part which is supplied by ACA perforators or the rHA) and the rostral tip of the putamen; T2 the caudate body and anterior tail, and the anterior putamen (except for the rostral tip), and T3 the posterior caudate tail and posterior putamen. LS infarct patterns I–IV were defined as the territorial infarctions to be expected from the LS occlusion patterns I–IV (LS infarct I=infarction of all subterritories T1–T3; LS infarct II=infarction of T2 and T3; LS infarct III=infarction of T3; and LS infarct IV=no infarction in any subterritory).
Vascular occlusion and associated infarction patterns lenticulostriate (LS) occlusion and LS infarct I–III (see text for definitions of these patterns). 1st row: Schematic sketches of LS infarct patterns superimposed on (normal) axial T1 weighted images at the level of the caudate head (left), and of the dorsal putamen/caudate tail (right). 2nd row: Postinterventional diffusion weighted MR images (∼3 days after mechanical thrombectomy) of three sample patients at the same sectional levels. 3rd and 4th rows: Original DSA images of the same sample patients prior to (third row) and after (fourth row) successful mechanical thrombectomy.
Image assessment
LS occlusion scores of individual patients were derived from preinterventional CT angiography (CTA) or MR angiography (MRA) and interventional DSA, including initial (prior to MTE) and final DSA runs (depicting the reperfused LSAs after successful recanalization), with the raters (JFK and EB in consensus read) blinded to infarction patterns and clinical data. In cases of discrepancies between preinterventional CTA/MRA and initial DSA runs (eg, due to thrombus migration), classification was based on CTA/MRA. In 71 patients referred to us from other hospitals, preinterventional images had been viewed by the respective interventionalist prior to treatment (eg, via teleradiological services), but were no longer accessible. In these cases, LS occlusion assignment was based on DSA images alone. LS infarct patterns were derived from postinterventional NECT or diffusion weighted MR images, and assigned to one of these patterns. This was done independently by two raters (JK and JFK), blinded to the LS occlusion patterns and clinical data. For cases with discrepant judgments, a blinded consensus read was performed, yielding a unified dataset for subsequent analyses.
Comparison of LS occlusion and LS infarct patterns
We then evaluated how well the vascular occlusion/LS occlusion patterns predicted the ensuing infarction/LS infarct patterns. For each patient and each of the predefined LS territories T1–T3, we checked whether the assigned LS occlusion pattern was ‘true’ or ‘false positive’, or ‘true’ or ‘false negative’. For instance, patients with LS occlusion I and LS infarct I (involvement of all MCA LSAs, infarction of all territories T1–T3), and patients with LS occlusion IV and LS infarct IV (occlusion distal to the LS perforators, no LS infarction), would score ‘true positive’ and ‘true negative’, respectively, for all territories T1–T3. Of note, within this scheme one patient could score both ‘false’ ratings for one and ‘true’ ratings for the other territories or vice versa. For example, a patient with LS occlusion III and LS infarct I (occlusion involving the lateral LSAs only, but infarction of all territories T1–T3), would score ‘false negative’ for T1 and T2, and ‘true positive’ for T3. From these data, we computed sensitivities, specificities, and positive and negative predictive values (PPV, NPV), separately for each LS territory, and as global estimates, for the sum of these parameters added up across the LS territories.
Finally, in patients with ‘false positive’ or ‘false negative’ ratings, interventional DSA was reassessed to identify any deviation from ‘normal’ vascular anatomy that might explain the respective discordancy. This re-evaluation had to be done unblinded to the respective LS infarct pattern, but was not used to modify the data from the initial blinded assessments.
Statistical analyses
Concordance analyses were based on contingency tables and Cohen's kappa (κ). Groupwise comparisons were based on χ² or, if applicable, Fisher's exact tests for categorical variables, and t tests for (normally distributed) continuous variables, as applicable. All statistical analyses were performed with SPSS release 23.0 (IBM corporation).
Results
A total of 279 patients (mean age 70.3±15.2 years, range 18–96 years, 148 (53%) women) fulfilled the inclusion criteria; 176/279 patients (63%) received IV rtPA bridging. Stent retrievers were used in most cases (245), in 9 cases in conjunction with direct aspiration techniques/ADAPT and in 10 cases in conjunction with older devices (eg, MERCI). Five patients were treated with ADAPT alone. Overall, up to date equipment (stent retrievers or ADAPT) was used in 250 procedures. Twenty-two patients were treated with older devices alone, 4 patients had inaccessible target vessels, and in 3 patients the target vessels were already recanalized at the time of the first diagnostic DSA run. Eight patients received a permanent stent in the MCA. Successful recanalization was achieved in 222/279 patients (79.6%), and in 207/250 patients (82.8%) treated with up to date equipment. The average symptom onset to successful recanalization time (SOR-t) (if applicable) was 4 h 25 min (IQR 3 h 20 min–5 h 5 min).
Vascular occlusion and infarct patterns
Inter-rater concordance regarding LS occlusion patterns was high (Cohen's κ=0.81). In the initial blinded assessment, 47 patients were assigned LS occlusion I, 89 patients LS occlusion II, 45 patients LS occlusion III, and 98 patients LS occlusion IV. LS infarct scores could not be assigned to five patients, in whom extensive hemorrhage or perilesional edema precluded adequate assessment, leaving 274 patients for further analyses. Eight patients showed infarction in ‘proximal’ territories (ie, T1 and/or T2) but sparing of the ‘distal’ territory T3 (posterior putamen and posterior caudate tail). These patients were assigned LS infarct scores according to the most ‘proximal’ territory involved, to be consistent with LS occlusion pattern definitions.
Generally, the observed infarctions corresponded well to the predefined LS infarct patterns. The two readers assigned identical scores in 247/274 cases, yielding an excellent concordance rate (k=0.86). Remarkably, concordance for the LS infarct scores derived from postinterventional CT (κ=0.82, n=144) was only slightly lower than for those derived from MRI (k=0.91, n=130).
LS occlusion and LS infarct scores were concordant in 191 cases (κ=0.59) (table 1). Most discordant scores (61/83, lower left half of table 1) were accounted for by ‘false negative’ LS occlusion patterns (actual infarction in an LS territory despite apparent sparing of the associated LS perforators). In contrast, ‘false positive’ patterns (no infarction in an LS territory despite apparent involvement of the associated LS perforators) were rare (22/83).
Contingency table assessing the concordance of lenticulostriate occlusion and lenticulostriate infarct patterns
This issue was assessed in detail by determining the relation between predicted infarction (as by the assigned LS occlusion pattern) and observed infarction (as by the assigned LS infarct pattern) separately for each of the three prespecified LS territories (table 2): ‘false positive’ LS occlusion patterns were rare for all territories, whereas ‘false negative’ patterns were more frequent, in particular in the ‘proximal’ territory I.
Assessment of the predictive performance of lenticulostriate occlusion patterns regarding infarction in the predefined LS territories, as by the assigned lenticulostriate infarct pattern
Consequently, LS occlusion patterns, as a predictor of ensuing infarction in the associated territories, had high specificities and PPVs (92.2% and 91.1%), but somewhat lower sensitivities and NPVs (79.6% and 83.3%) (table 2).
‘False positive’ LS occlusion patterns might indicate that imminent LS infarction had been prevented by MTE. However, PPVs were equally high in successfully recanalized patients (table 2). Recanalization rates (86.4% vs 79.4%, p=0.58), the use of IV rtPA bridging (72.7% vs 62.3%, p=0.37), use of up to date equipment (100.0% vs 90.3%, p=0.23), and SOR-ts (3 h 50 min±1 h 17 min vs 4 h 28 min±1 h 20 min, p=0.1, t test) were similar in patients with and without ‘false positive’ LS occlusion patterns. Thus successful MTE did not significantly alter the high rate of striatal infarction, and ‘false positive’ LS occlusion patterns could not generally be attributed to MTE or recanalization.
We therefore re-examined the DSA images of all 22 patients with ‘false positive’ LS occlusion patterns to identify any anatomical peculiarities. This revealed initially unnoted deviations from the standard vascular supply scheme that explained the ‘unexpected’ preservation of LS tissue in as many as 15 of these 22 patients. Three patterns could be distinguished (figure 2A–C): (i) groups of small perforating arteries arising from the most proximal M1 segment, the carotid T, or the ACA A1 segment, coursing far laterally to supply (parts of) the caudate body and tail, and the putamen (9 cases); (ii) an unusually prominent recurrent artery from the ACA A1 segment (analogous to a large Heubner artery), giving rise to perforators to the distal LS territories (3 cases); and (iii) a dense collateral network formed by small perforating arteries arising proximally, bypassing a (pre-existing, chronic) MCA stenosis, providing collateral flow to the distal MCA and ‘en passant’ supply to the LS territories (3 cases). The remaining seven patients without such variants had relatively short SOR-ts, but their range of SOR-ts still overlapped with the remainder of the sample (3 h 29 min±55 min, range 2 h 50 min–5 h 20 min, p=0.08). The seven ‘true’ false positive patterns could also not be attributed to too early imaging: five of these patients had postinterventional MRI, with the earliest scan acquired 75 h after symptom onset, and all five had obvious index stroke related infarctions elsewhere but not in the predicted LS territories. The earliest CT in the two remaining false positive cases was acquired 27 h after symptom onset.
Vascular supply variants in three sample patients who developed no (A, C) or only posterior (B) striatal infarctions despite proximal middle cerebral artery (MCA) M1 occlusions (large black or white arrowheads in A–C). (A) Sample patient No 1. Initial DSA before mechanical thrombectomy (MTE) shows a group of perforators (black arrows) from the most proximal M1 and the A1 segment, coursing far laterally to supply the caudate body, and tail and putamen. CT on day 7 after MTE (Thrombolysis in Cerebral Infarction (TICI) 3, symptom onset to recanalization time (SOR-t) 4 h 20 min) shows small ischemic lesions in the posterior insula and frontal operculum (white arrow), but no lenticulostriate (LS) infarction. (B) Sample patient No 2. Preinterventional CT angiography (uppermost/middle right) and initial DSA show a prominent recurrent artery from the distal A1 segment (small arrows), which courses laterally, giving rise to small arteries supplying the striatum (black arrows in DSA). Diffusion weighted MR images (DWI) (bottom panel) on day 3 after MTE (TICI 3 with ∼5 h 30 SOR-t) shows restricted infarction in the posterior putamen and caudate tail (LS infarct III). (C) Sample patient No 3. Initial DSA (upper left) shows a dense network of perforators arising from the most proximal M1 segment and the anterior cerebral artery A1 (black arrows), supplying the LS territories and providing collateral flow to the peripheral MCA (white arrows). DSA after one MTE maneuver (upper right) reveals presumably pre-existing high grade stenosis, which required permanent stenting because of rapid reocclusion (not shown). DWI on day 2 shows hemodynamic ‘watershed’ infarcts in the deep white matter (lower right), but no LS infarction.
Finally, we attempted to identify factors explaining the occurrence of ‘false negative’ LS occlusion patterns. Prior application of IV rtPA (60.7% vs 63.8%, p=0.66), employment of up to date equipment (90.7% vs 85.0%, p=0.17), recanalization rates (82.0% vs 77.5%, p=0.40), and SOR-ts (4 h 23 min±1 h 9 min vs 4 h 27 min±1 h 25 min, p=0.76, t test) were similar for patients with and without such patterns. Reassessment of DSA images revealed an anatomical variant in one patient with distal M1 occlusion (initial LS occlusion score IV), in whom a group of lateral LS arteries, originating from the initially occluded superior trunk of the postbifurcational M1 segment, provided supply to the proximal LS territory, possibly accounting for the complete striatal infarction (LS infarct I) observed. In the other 60 patients with ‘false negative’ patterns, no relevant vascular supply variants could be identified.
Discussion
These data show that patients with acute MCA M1 occlusion, and consecutive blockage of LSAs, will almost inevitably develop infarctions in the dependent LS territory, despite reasonably fast and technically effective recanalizing therapy. LSA involvement predicted infarction in associated LS territories with a PPV >90%. In most cases that seemed to deviate from this rule, closer inspection revealed anatomical variants that effectively explained the preservation of LS tissue. If these were taken into account, the specificity of LS occlusion patterns as predictors of ensuing LS infarctions would rise to 96.9%, and PPV to 96.3%.
Current endovascular stroke therapy is irrefutably effective,21–25 but the association between MCA occlusion and ensuing striatal infarction is thus still similarly tight as in earlier lesion studies, conducted long before MTE, and even prior to systemic thrombolysis.6 ,8 However, this may not hold indefinitely. The average SOR-t in our entire sample (4 h 25 min) was similar to that in the recent large scale trials on MTE,21–25 but considerably shorter in the seven patients with unexplained ‘false positive’ LS occlusion patterns (3 h 29 min, p=0.08). This suggests that LS infarction may become preventable if transport and treatment times are reduced. However, many patients with similarly short SOR-ts still developed the LS infarctions predicted by their LS occlusion pattern. For example, for the 74 patients recanalized within 4 h, the PPV of LS occlusion patterns would still be 83%. This indicates that (i) a radical improvement of SOR-ts will be required to reliably avert striatal infarcts in M1 occlusion and (ii) the fate of the dependent LS territories depends on additional factors, not only on time. Elucidating these factors is likely to provide important insights regarding determinants of infarct growth in the hyperacute phase of stroke evolution.
Most of the ‘false negative’ LS occlusion patterns remain unexplained. Pertinent variations in vascular supply could be identified in 1 of 61 cases only. Microscopic variations in supply zones indiscernible by DSA are ruled out by histological studies showing very little overlap of LS perforator territories.7 Conceivably, initially more proximal clots may have dissolved partially and/or migrated distally prior to imaging. Alternative explanations may implicate secondary growth of the primary ischemic lesion due to, for example, inflammatory processes. For now, one has to conclude that LS infarctions are almost never smaller, but sometimes larger than predicted by LS artery involvement. Whether the latter is due to (potentially modifiable) secondary processes causing true infarct growth, or generally explicable by (probably unmodifiable) events that had occurred already in the early phase of the evolving stroke, is another issue warranting further research.
Proximal M1 occlusion is associated with worse clinical outcome than distal M1 occlusion,26 ,27 which has been attributed to lesioning of the IC27 by more proximal occlusions. However, proximal MCA occlusions can involve the IC anterior limb, but will not lead to (clinically more obvious) lesions of the pyramidal tract as it passes through the basal IC genu and posterior limb at the level of the thalamus: these regions are usually supplied via the anterior choroidal artery, the posterior communicating artery, or the posterior cerebral artery. M1 occlusion may lesion the pyramidal tract as it forms the IC dorsal genu and posterior limb (between the caudate body and putamen), but these regions are supplied by the most lateral LS arteries. Moreover, observations in our patients show that IC involvement is much less consistent than involvement of the striatum, presumably reflecting that white matter has greater ischemic tolerance than gray matter.28 That more proximal MCA occlusions regularly lead to more extensive striatal infarctions provides an alternative explanation for the reportedly poorer outcome26 ,27 of these patients.
A recent study examined early angiographic indicators of basal ganglia infarction to predict limited ‘success’ of MTE29 in MCA occlusion. However, such indicators are difficult to evaluate if the predicted infarctions are almost invariably present. To develop meaningful predictors of clinical outcomes, and concepts to further improve clinical results of MTE, it is important to understand what the current practice of MTE can and what it cannot achieve. The present findings suggest that, with currently achievable SOR-ts, gray matter without collateral supply will almost invariably progress to infarction. They also imply that the benefit21–25 of MTE in M1 occlusion is not based on salvaging the basal ganglia, but rather, mostly on salvaging ‘peripheral’ cortical and subcortical areas or fiber tracts in the deep white matter. This may suggest that distal M1 occlusions should be treated with MTE as consequently as proximal M1 occlusions, perhaps irrespective of early signs of basal ganglia infarction. On the other hand, clearcut signs of LS infarction prior to endovascular therapy have been associated with a higher risk of subsequent haemorrhage30 and, if present, should not be neglected in cases of proximal M1 occlusion considered for endovascular therapy. If the potentially increased risk of reperfusion hemorrhage regularly outweighs the chance to preserve tissue at risk distal to the LS territories in such cases is, again, unclear. Further research that also takes into account the different vulnerability of gray and white matter is warranted to further elucidate basis and time dependency of clinical benefits, and risks, brought about by this extremely valuable therapy.
This was a retrospective study with associated limitations. First, images were in part based on postinterventional NECT, which may be less accurate than MRI. However, all relevant results were essentially identical, whether derived from NECT or MRI, indicating that this had no relevant impact on the data. Second, preinterventional images were no longer accessible in some cases. Occasionally, undocumented discrepancies to initial DSA might have been present that could explain some of the ‘false negative’ LS occlusion patterns. However, (i) preinterventional images were available for most patients, and in these such discrepancies were rare (26/203) and (ii) inaccessibility of preinterventional images was equal among patients with and without ‘false negative’ LS occlusion patterns (17/61 vs 54/213, p=0.69, χ²). Third, LS occlusion patterns were defined according to the beginning of the occluded segment, not the extension, because reliable information on the latter was not generally available. Eight patients exhibited infarction in ‘proximal’ LS territories (caudate head and body, rostral putamen), but sparing of the ‘distal’ territory (posterior caudate tail and posterior putamen), which could be explained by short segmental MCA occlusions not extending to the most lateral LS arteries. In two of these patients, suitable preinterventional CTAs were available, which indeed showed such patterns. This actually corroborates the tight relation between LS arterial occlusion and ensuing infarction, but we cannot provide according evidence for the remaining six patients. However, even if all of these were counted as additional cases of ‘false positive’ LS occlusion patterns, the overall PPV would decline only marginally to 88.4%. Finally, the assessment of infarctions from routine NECT or MRI is essentially qualitative. Subtle differences in the degree of LS tissue necrosis, depending on, for example, SOR-ts, may exist. Such differences might perhaps be revealed by rigorously timed quantitative diffusion weighted imaging, for example, and should be examined by pertinent research.
Conclusion
Striatal infarctions in MCA occlusion are determined by thrombus location and individual vascular anatomy. They cannot normally be averted by MTE, at least not with currently achievable symptom onset to recanalization times. However, there are exceptions to this rule. Elucidating the factors responsible for such exceptions may provide important insights into protective mechanisms and determinants of infarct growth in the hyperacute phase of large vessel occlusion stroke. LS infarctions, which are unaffected by the confounding influence of collateral flow to the involved territories, appear particularly suitable to study these issues.
References
Footnotes
Contributors JFK designed the research, acquired, analyzed, and interpreted the data, and wrote and revised the manuscript. JK acquired, analyzed, and interpreted the data, and revised the manuscript. EB acquired and analyzed the data, and revised the manuscript. CZ interpreted the data and revised the manuscript.
Competing interests None declared.
Ethics approval The study was approved by the institutional ethics committee.
Provenance and peer review Not commissioned; externally peer reviewed.