Abstract 16998: Gender and Age Specific Baseline Predictors of MACE in PEACE Trial Identified by Machine Learning

Introduction: The NHLBI supported Prevention of Events with Angiotensin-Converting Enzyme (ACE) Therapy trial (PEACE) (NCT00000558) found that the addition of ACE inhibitor trandolapril to conventional therapy in 8290 patients with stable coronary artery disease and preserved ejection fraction provided no benefit against MACE (cardiovascular death, nonfatal myocardial infarction, or the need for coronary revascularization), the composite primary endpoint. We reused publicly available individual patient-level PEACE data from NHLBI Data Repository (BioLINCC) to perform hypothesis-generating secondary analyses by machine learning (ML) using random survival forest (RSF) to identify gender and age group specific predictors for MACE. Methods: RSF was performed on 50 baseline variables for the MACE outcome in male and female and in age group (<60, 60-69, >69) cohorts. The top ten predictors identified in each cohort were included in a multivariate analysis using a Cox proportional hazards model with a multiple regression approach. Results: The top 10 predictors for the MACE selected by RSF are shown in Figure 1. Expected cardiovascular (CV) risk predictors like blood pressure, Canadian CV Society angina classification (CCS), age, and a history of various CV procedures consistently emerge amongst the top ten predictors of the primary MACE outcome across all gender and age specific subgroups. Interestingly, RSF also identified renal function biomarkers like serum potassium and glomerular filtration rate as common top ten predictors. Conclusion: Using ML, we uncovered in an unbiased fashion, gender and age groups specific unanticipated top predictors for MACE in PEACE trial. This underscores the value of gender and age specific predictors to examine the efficacy and outcomes of therapeutic interventions in advancing precision and personalized medicine.

Download Full-text

572. Real-world Effectiveness of COVID-19 mRNA Vaccines against Hospitalizations and Deaths in a Retrospective Cohort

Open Forum Infectious Diseases ◽

10.1093/ofid/ofab466.770 ◽

2021 ◽

Vol 8 (Supplement_1) ◽

pp. S388-S388

Author(s):

Farhaan S Vahidy ◽

Lauren Pischel ◽

Mauricio E Tano ◽

Alan Pan ◽

Marc L Boom ◽

...

Keyword(s):

Real World ◽

Retrospective Cohort ◽

Dose Group ◽

Proportional Hazards Model ◽

Cox Proportional Hazards ◽

Vaccine Effectiveness ◽

Cox Proportional Hazards Model ◽

Data Repository ◽

Area Deprivation ◽

Immunization Status

Abstract Background The effectiveness of Severe Acute Respiratory Syndrome Coronavirus 2 vaccines after two doses needs to be demonstrated beyond clinical trials. Methods In a retrospective cohort assembled from a cross-institution comprehensive data repository, established patients of the health care system were categorized as having received no doses, one dose or two doses of SARS-CoV-2 mRNA vaccine through April 4, 2021. Outcomes were COVID-19 related hospitalization and death. Results Of 94,018 patients 27.7% had completed two doses and 3.1% had completed one dose of a COVID-19 mRNA vaccine. The two dose group was older with more comorbidities. 1.0% of the two dose group had a COVID-19 hospitalization, compared to 4.0% and 2.7% in the one dose and no dose groups respectively. The adjusted Cox proportional-hazards model based vaccine effectiveness after two doses (vs. no dose) was 96%(95% confidence interval(CI):95–97), compared to 78%(95%CI:76–82) after one dose. After two doses, vaccine effectiveness for COVID-19 mortality was 97.9%(95%CI:91.7–99.5), and 53.5%(95%CI:0.28–80.8) after one dose. Vaccine effectiveness at preventing hospitalization was conserved across age, race, ethnicity, Area Deprivation Index and Charlson Comorbidity Indices. Cohort Enrollment and Distribution by Immunization Status and Vaccine effectiveness against mortality Cohort members are described by their immunization status and hospitalization at the end of the study period ending April 4th, 2021. Percentages compare this population to the total established patients. Each group is then divided into when hospitalized events occurred across immunization status. These percentages compare the number of events to the population in the immunization status at the end of the analysis period. Odds ratios for mortality were calculated and vaccine effectiveness calculated as 1 minus odds ratio times 100%. Conclusion In a large, diverse US cohort, receipt of two doses of an mRNA vaccine was highly effective in the real-world at preventing COVID-19 related hospitalizations and deaths with a substantive difference in effectiveness between one and two doses. Disclosures All Authors: No reported disclosures

Download Full-text

Abstract 15572: Gender, Race, and Age Specific Baseline Predictors of All-cause Mortality in STICHES Trial Identified by Machine Learning

Circulation ◽

10.1161/circ.142.suppl_3.15572 ◽

2020 ◽

Vol 142 (Suppl_3) ◽

Author(s):

Ian C Atkinson ◽

Anwar Husain ◽

Gauri Dandi ◽

Nuha Gani ◽

Zyannah Mallick ◽

...

Keyword(s):

Machine Learning ◽

Heart Failure ◽

Race And Ethnicity ◽

Smoking Status ◽

Left Ventricular ◽

Cox Proportional Hazards Model ◽

Data Repository ◽

Plasma Sodium ◽

Age Group ◽

All Cause Mortality

Introduction: NHLBI supported STICHES trial (The Surgical Treatment for Ischemic Heart Failure Extended Study) (NCT00023595) was conducted to test whether blood flow restoration by coronary revascularization recovers chronic left ventricular dysfunction and improves survival, as compared to medical therapy alone in patients with congestive heart failure and coronary artery disease amenable to surgical revascularization. We reused publicly available individual patient-level STICHES trial data from NHLBI Data Repository (BioLINCC) to perform hypothesis-generating secondary analyses by machine learning (ML) using random survival forest (RSF) to identify gender, race and ethnicity, and age specific predictors for all-cause mortality (ACM). Methods: The population was sub-grouped by gender (male vs. female), race (white vs. Hispanic/Latinos/non-white), and age (< 55, 55-60, 61-69, and ≥70). RSF was performed on 48 baseline variables from 1212 patients to identify predictors of ACM. Top 10 RSF predictors for each subgroup were included in a multivariate analysis using a Cox proportional hazards model. Results: Top 10 predictors of ACM are shown in Table 1. While known cardiometabolic and vascular predictors were among the top predictors, RSF uniquely identified renal function related biomarkers and plasma sodium among important top predictors across the subgroups. Age was an important predictor for male and female, Hispanics/Latinos/non-whites, and patient groups ≥70 years old. Also, top predictors of ACM were current smoking status among age groups of <55 and 55-60, clinical recruitment site in age group 61-69, and female gender in age group 55-60. Conclusions: Using ML, we uncovered in an unbiased fashion, gender, age, race and ethnicity specific, unanticipated top predictors of ACM in STICHES trial. This highlights the value of ML for analyzing disease and therapeutic intervention outcomes to help implement precision medicine.

Download Full-text

Abstract 16688: Effects of Evolocumab in Patients With Prior Percutaneous Coronary Intervention: An Analysis From the Fourier Trial

Circulation ◽

10.1161/circ.142.suppl_3.16688 ◽

2020 ◽

Vol 142 (Suppl_3) ◽

Author(s):

Remo H Furtado ◽

Antonio A Fagundes ◽

Kazuma Oyama ◽

Thomas A Zelniker ◽

Minao Tang ◽

...

Keyword(s):

Statin Therapy ◽

Proportional Hazards ◽

Proportional Hazards Model ◽

Absolute Risk ◽

Coronary Revascularization ◽

Cox Proportional Hazards ◽

Cox Proportional Hazards Model ◽

Atherosclerotic Cardiovascular Disease ◽

History Of ◽

Coronary Events

Introduction: Among patients with atherosclerotic cardiovascular disease (ASCVD), those with history of PCI represent an important population for potential high risk for cardiovascular (CV) events. We examined the clinical efficacy of the PCSK9 inibitor evolocumab in patients with prior PCI. Methods: FOURIER randomized 27,564 patients with ASCVD on statin therapy to evolocumab or placebo with a median follow-up of 2.2 yrs. The primary end point (PEP) was the composite of CV death, MI, stroke, unstable angina, or coronary revascularization; major coronary events were the composite of coronary death, MI, or coronary revascularization. The risk of events in patients with and without a history of PCI were compared in the placebo arm. The clinical benefit of evolocumab vs. placebo was compared using a Cox proportional hazards model. Results: 17,073 (62%) patients had prior PCI at baseline. Among patients in the placebo arm, those with prior PCI (N=8563) had a 1.6x higher rate of the PEP (16.8 vs 10.7%; adjusted HR 1.61; 95% CI 1.42-1.84 P<0.0001) and nearly double the rate of major coronary events (14.5 vs. 7.8%; P<0.0001; adjusted HR 1.72; 95% CI 1.49-1.99; Figure left). In patients with prior PCI, evolocumab reduced the risk of the PEP by 16% (HR 0.84; 95% CI 0.77-0.91; P<0.0001) and of major coronary events by 18% (HR 0.82; 95% CI 0.75-0.90, P<0.0001; Figure right), including a 30% reduction in fatal or non-fatal MI (P<0.001) and a 24% reduction in coronary revascularization (P<0.001). After the first year, there was a 25% reduction in major coronary events (HR 0.75, 95% CI 0.66-0.86, P<0.0001). The absolute risk reduction at 3 years with evolocumab for major coronary events was 2.8% in patients with prior PCI vs. 0.3% in those without. Conclusions: In a contemporary cohort with ASCVD on statin therapy, patients with prior PCI were at heightened risk for coronary events. Evolocumab was highly effective in this group, reducing major coronary events by 18% with a NNT at 3 years of only 36.

Download Full-text

Abstract 16309: Age Specific Baseline Predictors of All-cause Mortality in Systolic Blood Pressure Intervention Trial (SPRINT) Identified by Machine Learning

Circulation ◽

10.1161/circ.142.suppl_3.16309 ◽

2020 ◽

Vol 142 (Suppl_3) ◽

Author(s):

Nuha Gani ◽

Anwar Husain ◽

Gauri Dandi ◽

Ian Atkinson ◽

Zyannah Mallick ◽

...

Keyword(s):

Machine Learning ◽

Blood Pressure ◽

Systolic Blood Pressure ◽

Age Groups ◽

Cox Proportional Hazards Model ◽

Data Repository ◽

Diabetic Patients ◽

Age Group ◽

Intervention Trial ◽

All Cause Mortality

Introduction: The NHLBI supported Systolic Blood Pressure (SBP) Intervention Trial (SPRINT) (NCT01206062) aimed to identify an SBP target to reduce incidence of cardiovascular (CV) morbidity and mortality in hypertensive, non-diabetic patients of age ≥ 50 at increased CV risk. It found that intensive treatment (SBP target <120 mmHg) led to fewer major CV events and death but higher rates of adverse events. We reused publicly available patient-level SPRINT data from NHLBI Data Repository (BioLINCC) to perform hypothesis-generating secondary analyses by machine learning (ML) using random survival forest (RSF), to identify age specific baseline (bl) predictors for all-cause mortality (ACM). Methods: RSF was performed on 30 bl variables from 9361 patients in age group specific cohorts (50-59, 60-69, 70-79, 80-90). The identified top 10 predictors from each cohort were included in a multivariate analysis using a Cox proportional hazards model. Results: The top 10 predictors of ACM for age specific subgroups are shown in Figure 1. As expected, cardiovascular disease (CVD) predictors were selected, yet RSF distinctively identified renal biomarkers as important predictors, consistent with our previous analyses. Smoking status and history of CVD ranked as top predictors among age groups 50-59, 60-69, and 70-79. RSF also identified social factors, including race among age groups 60-69 and 80-90 and female gender among age groups 50-59 and 80-90 as important predictors for ACM. Lipid markers and medications used also showed up as top predictors. Specifically, polypharmacy emerged as a top predictor in age groups 60-69, 70-79, and 80-90, notably ranking higher in the 80-90 age group. Conclusions: Using ML, we uncovered in an unbiased fashion, unanticipated age specific top predictors for ACM in SPRINT trial. This highlights the value of ML for analyzing disease and therapeutic intervention outcomes and age specific prognostic factors to advance precision medicine.

Download Full-text

Abstract 15971: Gender, Race, and Age Specific Baseline Predictors of All-Cause Mortality in BARI2D Trial Identified by Machine Learning

Circulation ◽

10.1161/circ.142.suppl_3.15971 ◽

2020 ◽

Vol 142 (Suppl_3) ◽

Author(s):

Zyannah Mallick ◽

Nayab Mahmood ◽

Gauri Dandi ◽

Nowreen Haq ◽

Avantika Banerjee ◽

...

Keyword(s):

Machine Learning ◽

Type 2 Diabetes ◽

Medical Therapy ◽

Proportional Hazards Model ◽

Age Groups ◽

Cox Proportional Hazards Model ◽

Data Repository ◽

All Cause Mortality ◽

Hispanic White

Introduction: NHLBI supported Bypass Angioplasty Revascularization Investigation in Type 2 Diabetes trial (BARI2D) (NCT00006305) evaluated patients with type 2 diabetes and coronary artery disease. Primary trial analysis found no significant differences in rates of all-cause mortality (ACM) among patients who underwent 1) prompt revascularization with medical therapy versus aggressive medical therapy alone and 2) insulin-sensitization medical strategies versus insulin-provision. We reused publicly available individual patient-level data from NHLBI Data Repository (BioLINCC) to perform hypothesis-generating secondary analysis by machine learning (ML), using random survival forest (RSF) to identify gender, race, and age specific baseline predictors for ACM. Methods: The total 2368 trial participants was separated into several subgroups based on gender (female and male), age (40-49, 50-59, 60-69, 70-80), and race (Non-Hispanic White, Hispanic White, Non-Hispanic Non-White, and Hispanic Non-White). RSF was performed on 84 baseline variables to identify predictors of the primary outcome, ACM. The top 10 predictors for each subgroup were tested in a Cox proportional hazards model Results: Top 10 predictors of ACM are shown in Table 1. Although anticipated cardiovascular (CV) and diabetic predictors appeared among the top predictors, at the same time, renal function biomarkers like serum creatinine, urine albumin/creatinine ratio, and serum potassium uniquely showed among the top 5 predictors across the gender, age, and race specific subgroups. Conclusions: Using ML, we uncovered in an unbiased fashion, gender, race and age groups specific unanticipated top baseline predictors of ACM in BARI2D trial. This highlights the value of gender, race and age groups specific predictors of outcomes for determining the efficacy of therapeutic interventions and help advance precision medicine.

Download Full-text

Age at Exposure to Parental Suicide and the Subsequent Risk of Suicide in Young People

Crisis ◽

10.1027/0227-5910/a000468 ◽

2018 ◽

Vol 39 (1) ◽

pp. 27-36 ◽

Cited By ~ 5

Author(s):

Kuan-Ying Lee ◽

Chung-Yi Li ◽

Kun-Chia Chang ◽

Tsung-Hsueh Lu ◽

Ying-Yeh Chen

Keyword(s):

Young People ◽

Proportional Hazards ◽

Proportional Hazards Model ◽

Cox Proportional Hazards ◽

Cox Proportional Hazards Model ◽

Birth Registry ◽

Data Set ◽

Parental Suicide ◽

The Impact ◽

Age At Exposure

Abstract. Background: We investigated the age at exposure to parental suicide and the risk of subsequent suicide completion in young people. The impact of parental and offspring sex was also examined. Method: Using a cohort study design, we linked Taiwan's Birth Registry (1978–1997) with Taiwan's Death Registry (1985–2009) and identified 40,249 children who had experienced maternal suicide (n = 14,431), paternal suicide (n = 26,887), or the suicide of both parents (n = 281). Each exposed child was matched to 10 children of the same sex and birth year whose parents were still alive. This yielded a total of 398,081 children for our non-exposed cohort. A Cox proportional hazards model was used to compare the suicide risk of the exposed and non-exposed groups. Results: Compared with the non-exposed group, offspring who were exposed to parental suicide were 3.91 times (95% confidence interval [CI] = 3.10–4.92 more likely to die by suicide after adjusting for baseline characteristics. The risk of suicide seemed to be lower in older male offspring (HR = 3.94, 95% CI = 2.57–6.06), but higher in older female offspring (HR = 5.30, 95% CI = 3.05–9.22). Stratified analyses based on parental sex revealed similar patterns as the combined analysis. Limitations: As only register-based data were used, we were not able to explore the impact of variables not contained in the data set, such as the role of mental illness. Conclusion: Our findings suggest a prominent elevation in the risk of suicide among offspring who lost their parents to suicide. The risk elevation differed according to the sex of the afflicted offspring as well as to their age at exposure.

Download Full-text

Proliferation-dominant high-grade astrocytoma: survival benefit associated with extensive resection of FLAIR abnormality region

Journal of Neurosurgery ◽

10.3171/2018.12.jns182775 ◽

2020 ◽

Vol 132 (4) ◽

pp. 998-1005 ◽

Cited By ~ 4

Author(s):

Haihui Jiang ◽

Yong Cui ◽

Xiang Liu ◽

Xiaohui Ren ◽

Mingxiao Li ◽

...

Keyword(s):

Survival Benefit ◽

Proportional Hazards ◽

Proportional Hazards Model ◽

Characteristic Curve ◽

Extent Of Resection ◽

Cox Proportional Hazards ◽

Cox Proportional Hazards Model ◽

High Grade ◽

Extensive Resection ◽

High Grade Astrocytoma

OBJECTIVEThe aim of this study was to investigate the relationship between extent of resection (EOR) and survival in terms of clinical, molecular, and radiological factors in high-grade astrocytoma (HGA).METHODSClinical and radiological data from 585 cases of molecularly defined HGA were reviewed. In each case, the EOR was evaluated twice: once according to contrast-enhanced T1-weighted images (CE-T1WI) and once according to fluid attenuated inversion recovery (FLAIR) images. The ratio of the volume of the region of abnormality in CE-T1WI to that in FLAIR images (VFLAIR/VCE-T1WI) was calculated and a receiver operating characteristic curve was used to determine the optimal cutoff value for that ratio. Univariate and multivariate analyses were performed to identify the prognostic value of each factor.RESULTSBoth the EOR evaluated from CE-T1WI and the EOR evaluated from FLAIR could divide the whole cohort into 4 subgroups with different survival outcomes (p < 0.001). Cases were stratified into 2 subtypes based on VFLAIR/VCE-T1WIwith a cutoff of 10: a proliferation-dominant subtype and a diffusion-dominant subtype. Kaplan-Meier analysis showed a significant survival advantage for the proliferation-dominant subtype (p < 0.0001). The prognostic implication has been further confirmed in the Cox proportional hazards model (HR 1.105, 95% CI 1.078–1.134, p < 0.0001). The survival of patients with proliferation-dominant HGA was significantly prolonged in association with extensive resection of the FLAIR abnormality region beyond contrast-enhancing tumor (p = 0.03), while no survival benefit was observed in association with the extensive resection in the diffusion-dominant subtype (p=0.86).CONCLUSIONSVFLAIR/VCE-T1WIis an important classifier that could divide the HGA into 2 subtypes with distinct invasive features. Patients with proliferation-dominant HGA can benefit from extensive resection of the FLAIR abnormality region, which provides the theoretical basis for a personalized resection strategy.

Download Full-text

Association of time to surgery with leg pain after lumbar discectomy: is delayed surgery detrimental?

Journal of Neurosurgery Spine ◽

10.3171/2019.8.spine19613 ◽

2020 ◽

Vol 32 (2) ◽

pp. 160-167 ◽

Cited By ~ 2

Author(s):

Alessandro Siccoli ◽

Victor E. Staartjes ◽

Marlies P. de Wispelaere ◽

Marc L. Schröder

Keyword(s):

Rating Scale ◽

Proportional Hazards Model ◽

Cox Proportional Hazards ◽

Lumbar Discectomy ◽

Cox Proportional Hazards Model ◽

Neurological Deficits ◽

Leg Pain ◽

Specific Method ◽

Time To Surgery ◽

Patient Reported

OBJECTIVEWhile it has been established that lumbar discectomy should only be performed after a certain waiting period unless neurological deficits are present, little is known about the association of late surgery with outcome. Using data from a prospective registry, the authors aimed to quantify the association of time to surgery (TTS) with leg pain outcome after lumbar discectomy and to identify a maximum TTS cutoff anchored to the minimum clinically important difference (MCID).METHODSTTS was defined as the time from the onset of leg pain caused by radiculopathy to the time of surgery in weeks. MCID was defined as a minimum 30% reduction in the numeric rating scale score for leg pain from baseline to 12 months. A Cox proportional hazards model was utilized to quantify the association of TTS with MCID. Maximum TTS cutoffs were derived both quantitatively, anchored to the area under the curve (AUC), and qualitatively, based on cutoff-specific MCID rates.RESULTSFrom a prospective registry, 372 patients who had undergone first-time tubular microdiscectomy were identified; 308 of these patients (83%) obtained an MCID. Attaining an MCID was associated with a shorter TTS (HR 0.718, 95% CI 0.546–0.945, p = 0.018). Effect size was preserved after adjustment for potential confounders. The optimal maximum TTS was estimated at 23.5 weeks based on the AUC, while the cutoff-specific method suggested 24 weeks. Discectomy after this cutoff starts to yield MCID rates under 80%. The 24-week cutoff also coincided with the time point after which the specificity for MCID first drops below 50% and after which the negative predictive value for nonattainment of MCID first surpasses ≥ 20%.CONCLUSIONSThe study findings suggest that late lumbar discectomy is linked with poorer patient-reported outcomes and that—in accordance with the literature—a maximum TTS of 6 months should be aimed for.

Download Full-text

Survival of people living with AIDS with or without tuberculosis in Sao Paulo, Brazil

European Journal of Public Health ◽

10.1093/eurpub/ckaa166.835 ◽

2020 ◽

Vol 30 (Supplement_5) ◽

Author(s):

M V Tancredi ◽

S Sakabe ◽

C S B Domingues ◽

G F M Pereira² ◽

E A Waldman

Keyword(s):

Proportional Hazards Model ◽

Universal Access ◽

Sao Paulo ◽

Cox Proportional Hazards ◽

Cox Proportional Hazards Model ◽

Capital City ◽

São Paulo ◽

Aids Patients ◽

Probability Of Survival ◽

Aids Diagnosis

Abstract Background To estimate median survival time of AIDS patients, with and without tuberculosis (TB), in a cohort in Sao Paulo, Brazil, and to investigate survival predictors. Methods Retrospective cohort study of AIDS patients above 12 years old, registered at the Ministry of Health AIDS surveillance system between 2003-2007, and followed until 2014. Survival analysis used the Kaplan-Meier method and Cox proportional hazards model to estimate hazard ratios (HR), with respective 95% confidence intervals (CI = 95%). Results 35,515 patients were included, being 4,581 (12.9%) co-infected with TB. Among the latter, probability of survival 12 years after AIDS diagnosis was 95.2%, 82.9%, and 21.9%, respectively for patients receiving at least one third line ARV (HAART2), receiving triple therapy (HAART1) and the last one not on ARV. In the same period, the probability of survival for patients without TB, in the same order as for the therapeutic regimens, was 95.2%, 90.5%, and 40.9%, respectively. The main factors associated with survival, adjusted for the year of diagnosis, were: Living in the city of Sao Paulo (HR = 1,16;IC95% 1,01-1,32), living away from the capital city (HR = 1.43; 95%CI 1.25-1.62); or on the coast (HR = 1.49; 95%CI 1.21-1.82); having TB (HR = 1.70; 95%CI 1.49-1.87); above 49 years old (HR = 1.35; 95%CI 1.18-1.54); black (HR = 1.27; 95%CI 1.12-1.45); IV drug use (HR = 1.73; 95%CI 1.49-2.02); CD4+ below 200 cell/mm³ at AIDS diagnosis (HR = 2.31; 95%CI 1.97-2.72); viral load above 500 copies at AIDS diagnosis (HR = 1.99; 95%CI 1.72-2.30); HAART1 scheme (HR = 1.94; 95%CI 1.47-2.55); no ARV (HR = 8.22; 95%CI 2.95-22.87). Conclusions A large proportion of patients did not receive ARVs or were late diagnosed with AIDS, especially those with TB, whose survival was shorter. Survival is heterogeneous in the state, being lower in regions with higher TB rates. The results point to the need for specific strategies for patients with TB-HIV co-infection. Key messages Tuberculosis is the main cause of death among HIV-infected people, being responsible for one third of deaths in this group and causing a great impact on the survival of this population. The Brazilian policy of universal access to ARV and treatment for TB has increased the survival of AIDS-TB from 22% to 95% and in patients without TB from 50% to 95% up to 12 years after diagnosis.

Download Full-text

Adapting the Default Weighted Survival Analysis Modelling Approach to Model IFRS 9 LGD

Risks ◽

10.3390/risks9060103 ◽

2021 ◽

Vol 9 (6) ◽

pp. 103

Author(s):

Morne Joubert ◽

Tanja Verster ◽

Helgard Raubenheimer ◽

Willem D. Schutte

Keyword(s):

Survival Analysis ◽

Financial Reporting ◽

Proportional Hazards ◽

Proportional Hazards Model ◽

Survival Curve ◽

Cox Proportional Hazards ◽

Cox Proportional Hazards Model ◽

Reporting Standard ◽

International Financial Reporting Standard ◽

Ifrs 9

Survival analysis is one of the techniques that could be used to predict loss given default (LGD) for regulatory capital (Basel) purposes. When using survival analysis to model LGD, a proposed methodology is the default weighted survival analysis (DWSA) method. This paper is aimed at adapting the DWSA method (used to model Basel LGD) to estimate the LGD for International Financial Reporting Standard (IFRS) 9 impairment requirements. The DWSA methodology allows for over recoveries, default weighting and negative cashflows. For IFRS 9, this methodology should be adapted, as the estimated LGD is a function of in the expected credit losses (ECL). Our proposed IFRS 9 LGD methodology makes use of survival analysis to estimate the LGD. The Cox proportional hazards model allows for a baseline survival curve to be adjusted to produce survival curves for different segments of the portfolio. The forward-looking LGD values are adjusted for different macro-economic scenarios and the ECL is calculated for each scenario. These ECL values are probability weighted to produce a final ECL estimate. We illustrate our proposed IFRS 9 LGD methodology and ECL estimation on a dataset from a retail portfolio of a South African bank.

Download Full-text