Factors Influencing Outcome and Treatment Effect in PROACT II
Background and Purpose— The PROACT II study demonstrated a significant benefit from treatment with intra-arterial pro-urokinase (r-proUK) in patients with middle cerebral artery occlusion treated within 6 hours of stroke onset. The purpose of the current study was to examine baseline factors to determine predictors of good outcome and response to treatment.
Methods— We selected from the baseline clinical, radiologic, and angiographic data variables that considered possibly related to outcome. A univariate analysis was performed to examine the association between these baseline factors and good outcome, defined as a modified Rankin scale score ≤2. A multivariate model then selected the most important variables independently influencing prognosis. A risk score for each patient was constructed on the basis of the patient’s individual values for each independent variable. Patients were stratified into risk quartiles based on their risk scores, and an odds ratio for each risk quartile was calculated. The treatment effects of each quartile were compared.
Results— In the univariate analysis, screening National Institutes of Health stroke scale (NIHSS) score and age were strongly associated with good outcome. The multivariate model selected age, NIHSS score, and CT hypodensity as the most important prognostic variables. Dividing patients into quartiles based on risk scores achieved a uniform gradient of probability of good outcomes. A trend toward benefit of r-proUK treatment was seen in all risk quartiles, and no differential treatment effect was observed across risk groups.
Conclusions— There was no evidence of differential effect of r-proUK across subgroups of patients stratified by risk.
Following the experience in coronary ischemia, thrombolytic therapy is becoming established as part of the acute treatment of ischemic stroke. On the basis primarily of the results of the National Institute of Neurological Disorders and Stroke (NINDS) trial,1 systemic tissue-type plasminogen activator (tPA) is routinely used in patients for treatment of acute ischemic stroke who present within 3 hours of the onset of their symptoms. Early experience with patients treated with intra-arterial (IA) thrombolysis suggested improved outcomes and recanalization rates with this approach.2–9 Two randomized, controlled studies have also evaluated the safety and efficacy of IA thrombolysis performed through a catheter directly into the body of the thrombus.10,11 Direct catheter administration allows a maximized dose of a thrombolytic agent in the vicinity of the clot while potentially reducing the risk of systemic bleeding. The Pro-Urokinase (r-proUK) for Acute Cerebral Thromboembolism (PROACT) studies demonstrated the benefit of IA therapy with r-proUK in improving outcome from stroke due to occlusion of the middle cerebral artery (MCA) treated within 6 hours of onset.10,11 In PROACT II, patients treated with IA r-proUK had a 15% absolute and a 58% relative increase in good outcome (defined as modified Rankin scale [mRS] score of ≤2 at 90 days) compared with controls. Intracerebral hemorrhage associated with neurologic deterioration occurred in 10% of r-proUK–treated patients and 2% of controls at 24 hours. We analyzed the PROACT II database to look for specific factors that predicted outcome and to assess whether patients at higher or lower risk, based on these factors, responded differentially to IA r-proUK.
PROACT II was a randomized trial that compared 9 mg of IA r-proUK plus low-dose IV heparin with heparin alone.11 In this study, 180 patients aged 18 to 85 years with an M1 or M2 MCA occlusion confirmed by angiography and within 6 hours of symptom onset were randomized in a 2:1 ratio to treatment versus controls. Table 1 lists inclusion and exclusion criteria for PROACT II. Between February 1996 and August 1998, 12 323 patients were screened at 54 centers, and 474 underwent angiography to find 180 patients entered into the trial. At randomization, patients were stratified by the National Institutes of Health stroke scale (NIHSS) into 3 strata: 4 to 10, 11 to 20, and 21 to 30. r-proUK was infused over 2 hours, and angiography was repeated at 2 hours to assess recanalization. The primary end point was an mRS score ≤2 at 90 days. CT and angiographic studies were reviewed by a central group of neuroradiologists. The database from PROACT II was interrogated by using univariate and multivariate techniques.
For the univariate analysis, 25 categorical and 10 quantitative variables that were considered possibly related to outcome were selected from pretreatment clinical, radiologic, and angiographic information obtained in the PROACT II study. A univariate analysis was first performed to assess the relation between individual variables and the primary outcome measure of an mRS score ≤2 and the extent to which the variable influenced the size of the treatment effect. These analyses were achieved with a logistic model that included terms for treatment and the baseline variable being considered; an additional product term (treatment times baseline variable) was then added to allow treatment effect (expressed as an odds ratio [OR]) to differ according to the status of the baseline variable. Quantitative variables were included in these analyses as binary factors dichotomized at the median; the exception was NIHSS score, which was trichotomized (4 to 10, 11 to 20, and 21 to 30) as dictated by the stratification specified in the study protocol.3 The Bonferroni correction12 was used when assessing the significance of relations to allow for multiple comparisons.
To investigate the general relation between “risk” and treatment effect in PROACT II, we built a multivariable logistic risk model by stepwise selection from the baseline variables considered in the univariate analysis. Because the overall result of PROACT II demonstrated a significant treatment effect, “treatment” was entered into the model initially, and then subsequent steps added baseline variables in order of their prognostic importance with respect to the primary outcome. An important feature of this approach was that it allowed for intercorrelation between prognostic factors, such that variables were selected on the basis of their independent contribution to prognosis as opposed to that derived from correlations with other influential variables.
The important prognostic variables selected by this model were then used to create a risk score for each patient on the basis of that patient’s individual values for the prognostic variables. The risk score was computed as the sum over the selected variables of the patient’s baseline status times the estimated regression coefficient (the β coefficient) for the variable. The patients were then stratified into risk quartiles based on risk scores, and an OR for treatment effect was calculated for each risk quartile. The Breslow-Day test was used to compare quartiles for differences in treatment effect.
Analyses were performed for all randomized subjects (intention-to-treat), unless otherwise specified. Results of the univariate analysis are summarized in Table 2. As might be anticipated, older age and the presence of many of the history items tended to reduce the likelihood of a good outcome, as did increasing baseline NIHSS score. Because we considered a total of 35 individual baseline variables, the Bonferroni correction required a probability value of 0.05/35, or 0.0014, for any 1 analysis to be considered “significant.” Only NIHSS stratum and age >68 reached statistical significance by this criterion and thus, showed strong evidence of an influence on outcome. History of cerebrovascular disease, diabetes, and hypertension attained probability values close to this level and thus, could only be considered to have shown a “trend” toward a relation with outcome.
All potential predictive factors were offered to the multivariate model. The OR associated with treatment, after adjustment for other important prognostic variables, was 2.49 (P=0.022), greater than the OR obtained from the original PROACT II analysis of 2.13 (P=0.043). After treatment effect, which was forced in, the model selected age >68, NIHSS stratum. and CT hypodensity >5.25 mL as the most important prognostic variables (Table 3). No other factors added important amounts of prognostic information. Trichotomizing quantitative variables or treating them as continuous variables did not appreciably change the results of the multivariable analysis.
Based on the individual values for the 3 most important prognostic variables defined in the multivariate model, a risk score was assigned to each patient. The group was then divided into 4 risk quartiles based on the risk score (Table 4). The effectiveness of this risk score in separating patients with inherently different prognoses can be seen in the gradient of risk in the control group, which ranged from only a 5% likelihood of achieving a good outcome (mRS score ≤2) in the high-risk quartile to a 56% likelihood of a good outcome in the lowest-risk quartile. Within each risk quartile, there were more good outcomes (mRS score ≤2) in those treated with r-proUK; however, as might be expected in these relatively small subgroups of patients, the differences did not reach statistical significance. Although there was some variability in the size of the observed treatment effect (expressed as an OR) over risk quartiles, the extent of this variability was quite consistent with the natural play of chance. The formal test of the homogeneity of treatment effect across subgroups was firmly nonsignificant at P=0.91. The same lack of differential treatment effect was observed when either intent-to-treat or treated-as-allocated groups were used.
The categorical and quantitative variables were also examined for their individual influence on treatment effect (ie, subgroup effects). Because of the limited size of PROACT II and the multiple comparisons, this must be considered an exploratory analysis rather than definitive. Etiology of stroke was the only baseline variable close to significance in this analysis after correction for multiple comparisons (P=0.0090). Treatment with r-proUK was associated with a greater treatment effect with internal carotid artery (ICA) atherosclerosis or a cardioembolic etiology of stroke than with stroke due to unknown or other causes. Treatment effect tended to be greater in the absence of a hyperdense middle cerebral artery (HMCA) sign on CT; however, this effect was less convincing (P=0.030; Table 5).
The results of PROACT II indicated a clinically and statistically important improvement in the outcome of acute stroke patients when treated early with IA r-proUK. The effect of IA r-proUK was to more than double (OR=2.13) the odds of a good outcome compared with controls. The major conclusion from this secondary analysis of the PROACT II data is that this proportional treatment effect is quite consistent over subgroups of PROACT II patients at inherently different risk, as predicted from baseline factors. In the risk stratum with the worst prognosis for good outcome, only 17% of those treated with r-proUK reached the primary endpoint of an mRS score ≤2, but this was considerably more than the 5% in those who did not receive treatment. In the lowest-risk quartile, 68% of patients experienced a good outcome after r-proUK compared with 56% of the controls in this stratum. Comparing treatment effect across risk strata differs from the test-of-treatment effect used by the NINDS investigators.13 The NINDS intravenous tPA trial included >600 patients, which created larger subgroups for comparison of treatment effect. The small number of total patients and the 2:1 randomization limit the statistical power of comparing treatment effects across individual subgroups. Although it is biologically conceivable that treatment effect size would vary by individual baseline prognostic variables, the combined influence of all of these variables on “risk” might be the determining factor. By creating risk strata from predictors of risk, as determined by multivariate analysis, we avoid the problem of comparison of multiple small subgroups and increase the statistical power of detecting differences of treatment. Showing that the size of the relative treatment effect was independent of the level of risk allows clinicians to more confidently apply the results of PROACT II to their patients. For example, even in the lowest-risk stratum, with a 56% rate of good outcome in the control group, r-proUK might be expected to increase this rate to 68%, based on the common estimated treatment effect.
As noted previously, the OR (and probability value) associated with treatment after adjustment for other important prognostic variables in the multivariable analysis was 2.49 (P=0.022), stronger than the OR of 2.13 (P=0.043) obtained from the original intent-to-treat analysis. This indicates that randomization in PROACT II slightly favored controls in terms of inherent prognostic risk and that the trial might have slightly underestimated the true effect of treatment with r-proUK. Knowledge of prognostic variables such as those identified by this analysis might have implications for future thrombolytic trials that could stratify such variables at randomization to avoid imbalances affecting the primary results. This is particularly important for smaller studies like PROACT II, because the potential for bias diminishes as sample size increases.
We also examined treatment effect in individual subgroups, although because of the small number of patients, this must be considered an exploratory analysis to generate hypotheses for further testing in larger studies. Etiology of stroke showed a strong trend toward influencing the size of the treatment effect, with the greatest benefit in patients with stroke due to ICA atherosclerosis, followed by cardioembolic stroke. Stroke etiology was recorded by the individual investigators on the basis of interpretation of relevant clinical data. Standardized criteria such as TOAST14 or Oxfordshire classifications15 were not used. However, all investigators were experienced stroke neurologists. A differential response to treatment might relate to the composition or age of embolic material or to a smaller clot burden in emboli originating from ICA atherosclerosis compared with cardiac emboli or emboli from other sources.
Patients without the HMCA sign tended to have a greater response to r-proUK than those with this sign, although the magnitude of the effect was much less than for stroke etiology. This difference might also be explained by a large clot burden, different clot composition, or more proximal location of the clot in patients with the HMCA sign. Future studies of IA thrombolysis should focus on these possible factors that predict treatment effect to confirm the significance of this exploratory analysis and examine the physiological reasons for differences in treatment response.
Factors that influenced prognosis in this cohort of patients with an M1 or M2 MCA occlusions were similar to those in prior thrombolytic trials. In the NINDS trial of IV tPA administered within 3 hours of stroke onset, interactions between age and NIHSS scores, age and mean arterial pressure, and diabetes and early CT findings were independent predictors of good outcomes.13 No other prospective controlled trials of IA thrombolysis have been performed. One prior report of 76 consecutive patients treated with IA therapy identified NIHSS score, age, recanalization, etiology of infarct, and residual levels of blood flow by SPECT as independent factors influencing outcome.16 Other series of patients treated with IA thrombolysis found NIHSS score, early improvement, vessel recanalization, and length of occlusion to be factors associated with outcome.17,18 In a recent report of 100 patients with MCA occlusion treated with IA urokinase, age <60 years, low admission NIHSS score, and vessel recanalization were independently associated with excellent or good outcome, whereas diabetes was associated with poor outcome.19 Age and initial NIHSS score were also found to be independent predictors of outcome in our study. CT hypodensity size, which we found to be an important prognostic variable, was not reported in the study by Arnold et al.19 Prior reports of IA thrombolysis involved smaller numbers of patients and lacked controls for analysis of treatment effect. In addition, the occlusion sites were more varied, standard protocols were not applied, and multivariate analysis was not always performed. Even in our study, the relatively small numbers limit the strength of conclusions, particularly with regard to treatment effects.
Several factors, which in this analysis did not influence outcome or treatment effect, are noteworthy. Time to randomization and time to treatment were not strongly associated with outcome or treatment effect. The preponderance of patients treated after 5 hours in this study might have limited the ability to detect such a relation. Other thrombolytic trials identified time to treatment as a significant determinant of outcome.20 Abnormalities on initial CT, including evidence of edema and mass effect, also were not associated with outcome in this analysis, although a large area of hypodensity was one of the important prognostic variables selected by the logistic regression model. Analysis of CT findings from the NINDS tPA study also found no association between early CT abnormalities and outcome,21 although few patients had large areas of hypodensity at this early interval after stroke onset. In an analysis of factors associated with hemorrhage in the PROACT II study, Kase et al22 found that serum glucose was significantly associated with the risk of symptomatic hemorrhage. We found no association between diabetes or elevated glucose and outcome or treatment effect.
A trend toward greater treatment effect in patients with partial or complete collaterals was observed, but this did not reach statistical significance. This exploratory analysis does not exclude the possibility that collaterals maintain tissue viability and increase the response to treatment several hours after stroke onset, but further studies with larger numbers of patients will be needed to clarify this relation. Very few patients in this study had complete collaterals, the group that might be expected to have the greatest treatment response within a 6-hour time window.
The decision to treat an acute stroke patient with throm-bolysis would be greatly aided by information regarding the expected response. If a patient can be expected to have a robust response, other negative considerations might be overcome. Those less likely to respond might be excluded from treatment with a lower threshold. Our analysis indicates that a consistent pattern of treatment effect was seen over the range of baseline risk levels. Thus, when determining whether to treat an individual PROACT-like patient, the clinician can, with some confidence, apply the overall proportional treatment effect to the perceived risk of the patient in question.12
The PROACT II trial was funded by Abbott Laboratories. The authors would like to thank all the dedicated participating investigators in the PROACT II trial.
L.R. Wechsler, R. Roberts, A.J. Furlan, H.A. Rowley, and R.T. Higashida are consultants of Abbott Laboratories. C.M. Firszt and G.A. Schulz are employees of Abbott Laboratories.
- Received September 10, 2002.
- Revision received November 18, 2002.
- Accepted December 10, 2002.
Barnwell SL, Clark WM, Nguyen TT, O’Neill OR, Wynn ML, Coull BM. Safety and efficacy of delayed intraarterial urokinase therapy with mechanical clot disruption for thromboembolic stroke. Am J Neuroradiol. 1994; 15: 1817–1822.
Sasaki O, Takeuchi S, Koike T, Koizumi T, Tanaka R. Fibrinolytic therapy for acute embolic stroke: intravenous, intracarotid, and intra-arterial local approaches. Neurosurgery. 1995; 36: 246–252;discussion 252–253.
del Zoppo GJ, Ferbert A, Otis S, Bruckmann H, Hacke W, Zyroff J, Harker LA, Zeumer H. Local intra-arterial fibrinolytic therapy in acute carotid territory stroke: a pilot study. Stroke. 1988; 19: 307–313.
Suarez JI, Sunshine JL, Tarr R, Zaidat O, Selman WR, Kernich C, Landis DM. Predictors of clinical improvement, angiographic recanalization, and intracranial hemorrhage after intra-arterial thrombolysis for acute ischemic stroke. Stroke. 1999; 30: 2094–2100.
Edwards MT, Murphy MM, Geraghty JJ, Wulf JA, Konzen JP. Intra-arterial cerebral thrombolysis for acute ischemic stroke in a community hospital. Am J Neuroradiol. 1999; 20: 1682–1687.
del Zoppo GJ, Higashida RT, Furlan AJ, Pessin MS, Rowley HA, Gent M, PROACT. A phase II randomized trial of recombinant pro-urokinase by direct arterial delivery in acute middle cerebral artery stroke. Stroke. 1998; 29: 4–11.
Pagano M, Gauvreau K. Principles of Biostatistics. Pacific Grove, Calif: Duxbury; 2000: 292–293.
The NINDS t-PA Stroke Study Group. Generalized efficacy of t-PA for acute stroke: subgroup analysis of the NINDS t-PA stroke trial. Stroke. 1997; 28: 2119–2125.
Adams HP Jr, Bendixen BH, Kappelle LJ, Biller J, Love BB, Gordon DL, Marsh EE III. Classification of subtype of acute ischemic stroke: definitions for use in a multicenter clinical trial. Stroke. 1993; 24: 35–41.
Brandt T, von Kummer R, Muller-Kuppers M, Hacke W. Thrombolytic therapy of acute basilar artery occlusion: variables affecting recanalization and outcome. Stroke. 1996; 27: 875–881.
Gonner F, Remonda L, Mattle H, Sturzenegger M, Ozdoba C, Lovblad KO, Baumgartner R, Bassetti C, Schroth G. Local intra-arterial thrombolysis in acute ischemic stroke. Stroke. 1998; 29: 1894–1900.
Arnold M, Schroth G, Nedeltchev K, Loher T, Remonda L, Stepper F, Sturzenegger M, Mattle HP. Intra-arterial thrombolysis in 100 patients with acute stroke due to middle cerebral artery occlusion. Stroke. 2002; 33: 1828–1833.
Marler JR, Tilley BC, Lu M, Brott TG, Lyden PC, Grotta JC, Broderick JP, Levine SR, Frankel MP, Horowitz SH, et al. Early stroke treatment associated with better outcome: the NINDS rt-PA stroke study. Neurology. 2000; 55: 1649–1655.
Kase CS, Furlan AJ, Wechsler LR, Higashida RT, Rowley HA, Hart RG, Molinari GF, Frederick LS, Roberts HC, Gebel JM, et al. Cerebral hemorrhage after intra-arterial thrombolysis for ischemic stroke: the PROACT II trial. Neurology. 2001; 57: 1603–1610.