Can Machine Learning from Real-World Data Support Drug Treatment Decisions? A Prediction Modeling Case for Direct Oral Anticoagulants

Background: Decision making for the “best” treatment is particularly challenging in situations in which individual patient response to drugs can largely differ from average treatment effects. By estimating individual treatment effects (ITEs), we aimed to demonstrate how strokes, major bleeding events, and a composite of both could be reduced by model-assisted recommendations for a particular direct oral anticoagulant (DOAC). Methods: In German claims data for the calendar years 2014–2018, we selected 29 901 new users of the DOACs rivaroxaban and apixaban. Random forests considered binary events within 1 y to estimate ITEs under each DOAC according to the X-learner algorithm with 29 potential effect modifiers; treatment recommendations were based on these estimated ITEs. Model performance was evaluated by the c-for-benefit statistics, absolute risk reduction (ARR), and absolute risk difference (ARD) by trial emulation. Results: A significant proportion of patients would be recommended a different treatment option than they actually received. The stroke model significantly discriminated patients for higher benefit and thus indicated improved decisions by reduced outcomes (c-for-benefit: 0.56; 95% confidence interval [0.52; 0.60]). In the group with apixaban recommendation, the model also improved the composite endpoint (ARR: 1.69 % [0.39; 2.97]). In trial emulations, model-assisted recommendations significantly reduced the composite event rate (ARD: −0.78 % [−1.40; −0.03]). Conclusions: If prescribers are undecided about the potential benefits of different treatment options, ITEs can support decision making, especially if evidence is inconclusive, risk-benefit profiles of therapeutic alternatives differ significantly, and the patients’ complexity deviates from “typical” study populations. In the exemplary case for DOACs and potentially in other situations, the significant impact could also become practically relevant if recommendations were available in an automated way as part of decision making. Highlights It was possible to calculate individual treatment effects (ITEs) from routine claims data for rivaroxaban and apixaban, and the characteristics between the groups with recommendation for one or the other option differed significantly. ITEs resulted in recommendations that were significantly superior to usual (observed) treatment allocations in terms of absolute risk reduction, both separately for stroke and in the composite endpoint of stroke and major bleeding. When similar patients from routine data were selected (precision cohorts) for patients with a strong recommendation for one option or the other, those similar patients under the respective recommendation showed a significantly better prognosis compared with the alternative option. Many steps may still be needed on the way to clinical practice, but the principle of decision support developed from routine data may point the way toward future decision-making processes.

Reduced cardiovascular morbidity in patients with hemophilia: results of a 5-year multinational prospective study

Hemophilia is a congenital bleeding disorder caused by low clotting factor VIII or IX levels. Life expectancy of people with hemophilia (PWH) has increased with the availability of clotting factor concentrates. At the same time, the incidence of cardiovascular disease (CVD) has increased. In retrospective studies there are conflicting data if, despite this increase, the incidence is still lower than in the general population. We prospectively compared the incidence of CVD in PWH with the predicted incidence. This prospective, multicenter, observational study included adult PWH (>30y) from the Netherlands and United Kingdom (UK). They were followed for a 5-year period and CVD incidence was compared with a predicted event rate based on the QRISK2-2011 CVD risk model. The primary endpoint was the observed fatal and nonfatal CVD incidence after 5 years compared to the estimated events and in relation to severity of hemophilia. The study included 709 patients, of whom 687 (96.9%) completed 5 years follow up or reached an endpoint. For 108 patients the QRISK score could not be calculated at inclusion. For the remaining 579 fewer CVD events were observed than predicted: 9 versus 24 (RR 0.38; 95% CI: 0.18 - 0.80 p=0.01), corresponding with an absolute risk reduction of 2.4%. Severe hemophilia treated on demand had the highest risk reduction. There was no statistical significant relation between severity of hemophilia and incidence of CVD. In hemophilia a lower than predicted CVD incidence was found, supporting the theory that hemophilia protects against CVD. The study is registered at www.clinicaltrials.gov (identification number NCT01303900).

Antimicrobial Photodynamic Coatings Reduce the Microbial Burden on Environmental Surfaces in Public Transportation - a Field Study in Buses

Millions of people use public transportation daily worldwide and frequently touch surfaces, thereby producing a reservoir of microorganisms on surfaces increasing the risk of transmission. Constant occupation makes sufficient cleaning difficult to achieve. Thus, an autonomous, perma-nent antimicrobial coating (AMC) could keep down the microbial burden on such surfaces. A photodynamic AMC was applied to frequently touched surfaces in buses. The microbial burden (colony forming units, cfu) was determined weekly and compared to equivalent surfaces in buses without AMC (references). The microbial burden ranged from 0 – 209 cfu/cm² on references and from 0 – 54 cfu/cm² on AMC. The means were 13.4 ± 29.6 cfu/cm² on references and 4.5 ± 8.4 cfu/cm² on AMC (p<0.001). The difference of microbial burden on AMC and references was al-most constant throughout the study. Considering a hygiene benchmark of 5 cfu/cm², the data yield an absolute risk reduction of 22.6 % and a relative risk reduction of 50.7 %. In conclusion, photo-dynamic AMC kept down the microbial burden, reducing the risk of transmission of microor-ganisms. AMC permanently and autonomously contributes to hygienic conditions on surfaces in public transportation. Photodynamic AMC therefore are suitable for reducing the microbial load and closing hygiene gaps in public transportation.

P.150 Use of Antibacterial Envelopes for Prevention of Infection in Neuromodulation

Background: Neuromodulation unit placement carries a historic infection rate as high as 12%. TYRX antibacterial envelopes (Medtronic Inc., Minneapolis, MN), which are absorbable mesh envelopes that elute minocycline and rifampin, have been used in implantable cardiac devices with substantial risk reduction for infection. Methods: We conducted a retrospective cohort study of consecutive implantable pulse generator (IPG) and intrathecal pump unit implantation with a TYRX antibacterial envelope©. This cohort was then compared to a historical cohort of consecutive patients undergoing IPG or pump placement or revision prior to the use of the envelopes. Results: In the pre-envelope cohort of 151 IPGs in 116 patients, the infection rate was 18/151 (11.9%). In the antibacterial envelope cohort of 233 IPGs in 185 patients, the infection rate was (2.1%). The absolute risk reduction was 4.6% (95% CI, 0.045-0.048), The pre-envelope cohort of 41 pumps in 39 patients, the infection rate was 6/41 (14.6%). In the antibacterial envelope cohort of 59 pumps in 54 patients, the infection rate was (1.7%). The absolute risk reduction was 12.9% (95% CI 1.6-24.3). Conclusions: Usage of an antibacterial envelope for neuromodulation has resulted in a lower infection rate at our center. Based on these results, we recommend the use of antimicrobial envelopes.

Abstract 1122‐000024: Stereotactic IntraCerebral Underwater Blood Aspiration Improves Survival Following Intracerebral Hemorrhage as Compared to Predicted Mortality

Introduction : Intracerebral hemorrhage (ICH) is a devastating form of neurological injury with substantial mortality. Recent publications on minimal invasive surgery (MIS) for hematoma evacuation have suggested survival benefits in select patients. Since 2015, our center has been performing an MIS technique using continuous irrigation with aspiration through an endoscope (SCUBA). SCUBA does not require a stability scan and can be performed despite hematoma expansion, intraventricular hemorrhage or radiographic spot sign. We present the 30‐day mortality of our initial experience and compare it to predicted mortality by presenting ICH score. Methods : We performed a retrospective review of consecutively admitted patients with spontaneous non‐traumatic supratentorial ICH who underwent SCUBA between 12/2015 – 03/2019. The primary outcome was observed 30‐day mortality compared to predicted mortality by ICH score on presentation. Key secondary outcomes were operative markers, hospital length of stay, and discharge disposition. Results : One hundred and fifteen patients were identified, with mean (SD) ICH volume of 51.4mL (33.9mL) and median ICH score of 2. The median evacuation was 97% of the hematoma volume and 85% of patients had a residual clot burden of less than 15mL. Twelve patients died within one month of SCUBA for an overall mortality rate of 10.4%. This was significantly lower than the predicted mortality by ICH score of 35.1%, with an absolute risk reduction of 24.7%. When analyzed by presenting ICH score, significant mortality benefits were observed for all ICH scores > 2, with more pronounced differences at higher ICH scores (Table 1). Conclusions : This study suggests that MIS with the SCUBA technique for ICH may reduce predicted 30‐day mortality, with a number need to treat of 4 to prevent one mortality. Further evaluation of this technique in a randomized clinical trial is required.

Modeling the Clinical Implications of Andexanet Alfa in Factor Xa Inhibitor–Associated Intracerebral Hemorrhage

Background and Objectives:Andexanet alfa was recently approved as a reversal agent for the Factor Xa inhibitors (FXai) apixaban and rivaroxaban, but its impact on long-term outcomes in FXai-associated intracerebral hemorrhage (ICH) is unknown. We aimed to explore potential clinical implications of andexanet alfa in FXai-associated ICH in this simulation study.Methods:We simulated potential downstream implications of andexanet alfa across a range of possible hemostatic effects using data from a single center that treats FXai-associated ICH with prothrombin complex concentrate (PCC). We determined baseline probabilities of inadequate hemostasis across FXai and non-FXai patients via multivariable regression models, then determined probabilities of unfavorable 3-month outcome (modified Rankin Scale 4-6) using models comprising established predictors and each patient’s calculated probability of inadequate hemostasis. We applied bootstrapping with model parameters from this derivation cohort to simulate a range of hemostatic improvements and corresponding outcomes, then calculated absolute risk reduction (relative to PCC) and projected number needed to treat (NNT) to prevent one unfavorable outcome.Results:Training models using real-world patients (n=603 total; 55 FXai) had good accuracy in predicting inadequate hemostasis (AUC 0.78) and unfavorable outcome (AUC 0.78). Inadequate hemostasis was strongly associated with unfavorable outcome (OR 4.5, 95% CI 2.0-9.9) and occurred in 11.4% of FXai patients. Across simulated FXai patients comparable to those in the ANNEXA-4 study, predicted absolute risk reduction of unfavorable outcome was 4.9% (95% CI 1.3%-7.8%) when the probability of inadequate hemostasis was reduced by 33%, and 7.4% (95% CI 2.0%-11.9%) at 50% reduction, translating to projected NNTs of 21 (cumulative cost $519,750) and 14 ($346,500), respectively.Discussion:Even optimistic simulated hemostatic effects suggest that the costs and potential benefits of andexanet alfa should be carefully considered. Placebo-controlled randomized trials are needed before its use can definitively be recommended.

Ischemic Benefit and Hemorrhage Risk Of Ticagrelor-Aspirin Versus Aspirin In Patients With Acute Ischemic Stroke Or TIA

Background and Purpose: In patients with acute mild-moderate ischemic stroke or high-risk transient ischemic attack (TIA), the Acute Stroke or Transient Ischemic Attack Treated with Ticagrelor and Aspirin for Prevention of Stroke and Death (THALES) trial demonstrated that when added to aspirin, ticagrelor reduced stroke or death but increased risk of severe hemorrhage compared with placebo. The primary efficacy outcome of THALES included hemorrhagic stroke and death, events also counted in the primary safety outcome. We sought to disentangle risk and benefit, assess their relative impact, and attempt to identify subgroups with disproportionate risk or benefit. Methods: In a randomized, placebo-controlled, double-blind trial of patients with mild-to-moderate acute noncardioembolic ischemic stroke or high-risk TIA, patients were randomized within 24 hours after symptom onset to a 30-day regimen of either ticagrelor plus aspirin or matching placebo plus aspirin. For the present analyses, we defined the efficacy outcome, major ischemic events, as the composite of ischemic stroke or non-hemorrhagic death, and defined the safety outcome, major hemorrhage, as intracranial hemorrhage or hemorrhagic death. Net clinical impact was defined as the combination of these two endpoints. Results: In 11 016 patients (5523 ticagrelor-aspirin and 5493 aspirin), a major ischemic event occurred in 294 patients (5.3%) in the ticagrelor-aspirin group and in 359 patients (6.5%) in the aspirin group (absolute risk reduction 1.19%, 95%CI 0.31%-2.07%). Major hemorrhage occurred in 22 patients (0.4%) in the ticagrelor-aspirin group and 6 patients (0.1%) in the aspirin group (absolute risk increase 0.29%, 95% CI, 0.10-0.48%). Net clinical impact favored ticagrelor-aspirin (absolute risk reduction 0.97%, 95% CI, 0.08%-1.87%). Findings were similar when different thresholds for disability were applied and over a range of predefined subgroups. Conclusions: In patients with mild-moderate ischemic stroke or high-risk TIA, ischemic benefits of 30-day treatment with ticagrelor-aspirin outweigh risks of hemorrhage. Registration: URL: http://www.clinicaltrials.gov; Unique identifier: NCT03354429

Abstract 02: Distribution Of Predicted Cardiovascular And All-cause Mortality Benefit Of Intensive Vs Standard Blood Pressure Control Among Us Adults Eligible For The Systolic Blood Pressure Intervention Trial

Introduction: If resources are scarce, achieving national SBP control goals will require prioritizing treatment among those likely to benefit. To identify patients with greatest predicted benefit with intensive SBP treatment and estimate population sizes, we applied algorithms to community samples who met the SPRINT enrollment criteria. Methods: The published algorithms separately predict the absolute risk reduction in CVD events and mortality at 3.26 years with intensive (<120 mm Hg) vs standard (<140 mm Hg) SBP lowering. We applied and calibrated the algorithms to SPRINT standard arm participants (n=4 399) and samples meeting SPRINT enrollment criteria from the National Health and Nutrition Examination Survey (NHANES, n=1 297) and the Reasons for Geographic And Racial Differences in Stroke (REGARDS, n=2 785). Predicted absolute risk reduction estimated number needed to treat (NNT), categorized as <50, 50-100, and ≥100. Observed 3.26 year CVD event (SPRINT, REGARDS) and mortality rates (all cohorts) were calculated. Results: The median ages were 67 (SPRINT), 69 (NHANES), and 72 (REGARDS). Greater proportions of NHANES and REGARDS vs SPRINT had predicted NNT <100 for CVD events (NHANES 94.8%, REGARDS 99.2%, SPRINT 87.8%) and mortality (NHANES 64.3%, REGARDS 63.7%, SPRINT 38.8%) ( Table ). Event rates were comparable within NNT groups. Conclusions: Predicted NNT distributions differ between cohorts but event rates are similar. Most adults who meet SPRINT enrollment criteria have predicted NNT <100 for CVD and mortality with intensive SBP treatment. These results suggest that published algorithms can identify those most likely to benefit and can guide implementation.

Evaluation of an intervention targeted with predictive analytics to prevent readmissions in an integrated health system: observational study

Objectives To determine the associations between a care coordination intervention (the Transitions Program) targeted to patients after hospital discharge and 30 day readmission and mortality in a large, integrated healthcare system. Design Observational study. Setting 21 hospitals operated by Kaiser Permanente Northern California. Participants 1 539 285 eligible index hospital admissions corresponding to 739 040 unique patients from June 2010 to December 2018. 411 507 patients were discharged post-implementation of the Transitions Program; 80 424 (19.5%) of these patients were at medium or high predicted risk and were assigned to receive the intervention after discharge. Intervention Patients admitted to hospital were automatically assigned to be followed by the Transitions Program in the 30 days post-discharge if their predicted risk of 30 day readmission or mortality was greater than 25% on the basis of electronic health record data. Main outcome measures Non-elective hospital readmissions and all cause mortality in the 30 days after hospital discharge. Results Difference-in-differences estimates indicated that the intervention was associated with significantly reduced odds of 30 day non-elective readmission (adjusted odds ratio 0.91, 95% confidence interval 0.89 to 0.93; absolute risk reduction 95% confidence interval −2.5%, −3.1% to −2.0%) but not with the odds of 30 day post-discharge mortality (1.00, 0.95 to 1.04). Based on the regression discontinuity estimate, the association with readmission was of similar magnitude (absolute risk reduction −2.7%, −3.2% to −2.2%) among patients at medium risk near the risk threshold used for enrollment. However, the regression discontinuity estimate of the association with post-discharge mortality (−0.7% −1.4% to −0.0%) was significant and suggested benefit in this subgroup of patients. Conclusions In an integrated health system, the implementation of a comprehensive readmissions prevention intervention was associated with a reduction in 30 day readmission rates. Moreover, there was no association with 30 day post-discharge mortality, except among medium risk patients, where some evidence for benefit was found. Altogether, the study provides evidence to suggest the effectiveness of readmission prevention interventions in community settings, but further research might be required to confirm the findings beyond this setting.