598 Are risk scores sufficient to stratify patients undergoing lead extraction? A single-centre analysis

Abstract Aims Device implantation is growing exponentially, as well as associated infections, ranging from isolated pocket erosion to endocarditis and bacteraemia, all worsening the prognosis of patients with frailty and comorbidity. Transvenous lead extraction (TLE) can resolve the complications, although a 1-year mortality risk of up to 25% is reported; despite higher health costs, prolonged hospitalization, and poor quality of life, strategies for predicting increased infection risk and reduced infection incidence are yet missing. Currently applied clinical scores do not consider etiologic microbial agents. We aimed to assess whether PADIT and UPCM scores could be implemented when bacteria or fungi are known to be causative of infection, and how these agents affected the outcome. Methods and results A retrospective analysis of patients undergone cardiac implantable electronic device (CIED) pocket revision, and/or TLE between 2016 and 2021 was performed. For each procedure, microbiological samples of both generator pocket tissue and intracardiac portions of the leads were analysed. In addition, blood cultures were performed in three sets. Transesophageal echocardiography was performed in all cases for ruling out suspected endocarditis. Spearman ad Pearson coefficients were tested for correlation among microorganism, prior infection and/or procedure, PADIT and UPCM scores; a P-value less than 0.05 was considered significant. We analysed 14 patients (10 males, 4 females, mean age ± SD: 72 ± 13): one case (4%) affected by pocket erosion, seven cases (50%) affected by both pocket site and lead infection (with associated bacteraemia in one subject), and one case (4%) due to lead-related infective endocarditis. Of these, five (36%) underwent device replacement, while nine (64%) to extraction or pocket/lead revision. Nine (64%) patients had positive culture examinations (Figure 1). The correlation method gave a statistically significant association between Gram- infection and prior sepsis (r 0.63; P-value 0.02). We considered the number of procedures on the same pocket and/or CIED previous infections as markers of frailty and increased infectious risk. As expected, the PADIT score, but not UPCM, significantly correlated with the number of previous procedures (r 0.70; P-value 0.006). Indeed, both scores had a similar infectious risk prediction. Conclusions In our analysis, predictive PADIT score of infectious risk performed better than UPCM, while both proved their reliability in identifying high-risk patients. The absence of correlation between UPCM score and infective agents is not conclusive, but probably due to the small sample size. Interestingly, growing rate of device reinfection correlates with the risk of Gram- bacterial infection. Thus, the integration of the microbiological data in the current prediction models could significantly increase their performance.

Download Full-text

Prediction Models and Scores in Adult Congenital Heart Disease

Current Pharmaceutical Design ◽

10.2174/1381612827999210111181554 ◽

2021 ◽

Vol 27 ◽

Author(s):

Alexandra Arvanitaki ◽

Despoina Ntiloudi ◽

George Giannakoulas ◽

Konstantinos Dimopoulos

Keyword(s):

Congenital Heart Disease ◽

Heart Disease ◽

Congenital Heart ◽

Prediction Models ◽

Small Sample Size ◽

Adult Congenital Heart Disease ◽

Small Sample ◽

Risk Scores ◽

Post Intervention ◽

Size Heterogeneity

Abstract:: Nowadays, most patients with congenital heart disease survive to adulthood, thanks to advances in pediatric cardiac surgery, but often present with various comorbidities and long-term complications, posing challenges in their management. The development and clinical use of risk scores for the prediction of morbidity and/or mortality in adults with congenital heart disease (ACHD) is fundamental in achieving optimal management for these patients, including appropriate follow-up frequency, treatment escalation and timely referral for invasive procedures or heart transplantation. In comparison with other fields of cardiovascular medicine, there are relatively few studies that report prediction models developed in the ACHD population, given the small sample size, heterogeneity of the population and relatively low event rate. Some studies report risk scores originally developed in pediatric congenital or non-congenital population, externally validated in ACHD with variable success. Available risk scores are designed to predict heart failure or arrhythmic events, all-cause mortality, post-intervention outcomes, infective endocarditis or atherosclerosis-related cardiovascular disease in ACHD. A substantial number of these scores are derived from retrospective studies and are not internally or externally validated. Adequately validated risk scores can be invaluable in clinical practice and an important step towards personalized medicine. Multicenter collaboration, adequate study design and the potential use of artificial intelligence are important elements in the effort to develop reliable risk scores for the ACHD population.

Download Full-text

Retrospective, observational study on usage of evogliptin in T2DM patients: A real-World experience in Indian patients

Indian Journal of Pharmacy and Pharmacology ◽

10.18231/j.ijpp.2021.035 ◽

2021 ◽

Vol 8 (3) ◽

pp. 205-207

Author(s):

Abhijit Trailokya ◽

Amol Aiwale ◽

Roshan Pawar ◽

Suhas Erande

Keyword(s):

Real World ◽

Small Sample Size ◽

Small Sample ◽

Hypoglycemic Agents ◽

P Value ◽

Base Line ◽

Oral Hypoglycemic Agents ◽

Naive Patient ◽

Indian Patients ◽

Treatment Naïve

This study aimed to assess effectiveness and safety of Evogliptin 5 mg in patients with T2DM who were prescribed Evogliptin alone or with other oral hypoglycemic agents in real world scenario. Overall 20 patients who received Evogliptin as routine clinical practice in management of T2DM were analyzed retrospectively from single center. Data collected from past medical records. Primary endpoint was mean changes in HbA1c from baseline to weeks 24 and secondary endpoints were Change in HbA1c from baseline to weeks 12 Change from baseline in FPG & PPG at weeks 12 & 24.Significant reduction in HbA1c at the end of 12 and 24 weeks of Evogliptin therapy was - 0.9% and -1.45% respectively from the baseline of HbA1c 8.6% (p value <0.001). At the end of 12 and 24 weeks of addition of Evogliptin, significant reduction in FBG were seen i.e -49.5 mg/dl and -90.7mg/dl respectively from base line of 182 mg/dl and reduction in PPG was -79.4mg/dl and -116.6mg/dl respectively from base line 277 mg/dl (p value <0.001). Evogliptin was found to be effective when added to the patients who were uncontrolled on dual / triple oral anti-diabetic medications and even in treatment naïve patient. It effectively showed reduction in HbA1c, FBG and PPG and the end of 12 and 24 weeks when added to existing anti-diabetic medications & well tolerated in type 2 diabetes Indian patients.Small sample size and retrospective study

Download Full-text

Discrepancy of Blast Percentage between the Bone Marrow Aspirate and Flow Cytometry and Its Impact on Survival Outcomes in Patients with Myelodysplastic Syndromes Excess Blast (MDS-EB)

Blood ◽

10.1182/blood-2019-125439 ◽

2019 ◽

Vol 134 (Supplement_1) ◽

pp. 5441-5441

Author(s):

Meera Yogarajah ◽

Phuong L. Nguyen ◽

Rong He ◽

Hassan B. Alkhateeb ◽

Mithun Vinod Shah ◽

...

Keyword(s):

Bone Marrow ◽

Flow Cytometry ◽

Small Sample Size ◽

Scoring Systems ◽

Standard Of Care ◽

Risk Groups ◽

Small Sample ◽

International Prognostic Scoring System ◽

P Value ◽

Number Of Patients

Background MDS is a heterogeneous disease and the revised International Prognostic Scoring System (IPSS-R) is utilized in prognostication. The percentage (%) of blasts in the bone marrow is determined in the aspirate morphologically. Though the former is the standard of care the blast percentage is also reported by flow cytometry and biopsy which can many times be inconsistent. We previously presented the utilization of biopsy based blast percentage which showed meaningful prognostic groups compared to aspirate. In this study we compare the blasts as reported by the aspirate and flow cytometry in MDS-EB in calculating IPSS-R. Methods The MDS database was reviewed for cases of MDS-EB after due IRB approval at the Mayo clinic. We calculated IPSS-R scores based on the aspirate blast % (IPSS-RAsp) and flow blast% (IPSS-Rfl). The aspirate blast percentage was reported morphologically. Suboptimal aspirates were excluded from the study. The flow blast percentage was determined by immunophenotyping. The overall survival (OS) was determined by IPSS-RAsp and IPSS-RFl. OS estimates were calculated by Kaplan-Meier curves and log-rank testing using JMP v.13. Uno's concordance statistic was used to compare the 2 risk scoring systems. Results Of 1322 patients, 431 (33%) cases were identified with MDS-EB out of which 120 (29%) cases had blasts reported in the aspirate and flow. Based on aspirate MDS EB1: 54% (n=65), MDS EB2 46% (n=55). The hematological, cytogenetic and R-IPSS categories were compared between MDS-EB1 and MDS- EB 2. The blast percentage and hemoglobin levels was significantly different between MDS-EB1 and EB2 as seen in table 1, however the IPSS-R risk groups were not significantly different. The flow cytometry was concordant with aspirate in 66/120 (55%) cases. Out of the dis-concordant cases only 20% (11/54) was upstaged by flow cytometry with most of the patients being down staged as expected by the techniques used in processing the blood and hence not reliable when reported low (Figure 1). The OS outcomes based on the IPSS- R asp, IPSS-Rfl areshown in figure 2A,2B .The p value with aspirate based R-IPSS was more significant than flow cytometry based R-IPSS (p= 0.0007 vs 0.0174). We compared the two models for observed OS differences using the Uno model which was not statistically significant. (p= 0.6) Conclusions Both models did not show a difference which is likely due to the very small sample size. However flow cytometry did down stage more patients when disconcordant and may have less value in that setting. It would be ideal to compare all 3 models aspirate, biopsy and flow cytometry however we did not have enough number of patients to do the comparison. Disclosures Patnaik: Stem Line Pharmaceuticals.: Membership on an entity's Board of Directors or advisory committees. Al-Kali:Astex Pharmaceuticals, Inc.: Research Funding.

Download Full-text

Impact of antibiotic pretreatment on cultures in children with osteomyelitis and septic arthritis: a retrospective review

BMC Pediatrics ◽

10.1186/s12887-021-02806-w ◽

2021 ◽

Vol 21 (1) ◽

Author(s):

Amanda Lansell ◽

Yasasvi Vasili ◽

Parminder S. Suchdev ◽

Janet Figueroa ◽

Anjali Kirpalani

Keyword(s):

Septic Arthritis ◽

Pediatric Patients ◽

Small Sample Size ◽

Positive Culture ◽

Small Sample ◽

Blood Cultures ◽

Pathogen Identification ◽

Culture Positivity ◽

Antibiotic Duration ◽

Bone Cultures

Abstract Background In the management of pediatric osteomyelitis or septic arthritis, delay in treatment may affect outcome, while receipt of antibiotics prior to culture may affect culture results. We aimed to determine if pathogen identification decreased in cultures that were pretreated with antibiotics. Methods We conducted a retrospective cohort study of 584 hospitalized children between 30 days and 18 years of age admitted to two tertiary children’s hospitals. Logistic regression assessed the effect of antibiotic duration on blood, bone, joint aspirate, and “other” culture positivity. Results Overall, 42% of blood cultures, 70% of bone cultures, 39% of joint cultures, and 70% of “other” cultures were positive. Compared with children who did not receive antibiotics prior to culture, there were no significant differences in odds of a positive culture in children whose cultures were pretreated with antibiotics for any of the culture types [OR (95% CI) 0.90 (0.56–1.44) for blood cultures, 0.77 (0.25–2.34) for bone cultures, 0.71 (0.39–1.28) for joint cultures, 1.18 (0.58–2.41) “for other” cultures; all p > 0.05]. Furthermore, the duration (hours) of antibiotics in the pretreated cultures was also not a significant predictor of culture positivity (OR ranged from 0.99–1.00 for all cultures, p > 0.05). Conclusions Culture positivity was not associated with antibiotic pretreatment in any of the samples, even for longer duration of antibiotics prior to culture, though the small sample size of subgroups is an important limitation. In pediatric patients hospitalized with osteomyelitis and/or septic arthritis, early initiation of antibiotics may not affect culture positivity.

Download Full-text

Role of an extended pathology exam in risk classification of testicular germ cell tumors (GCTs).

Journal of Clinical Oncology ◽

10.1200/jco.2012.30.15_suppl.e15032 ◽

2012 ◽

Vol 30 (15_suppl) ◽

pp. e15032-e15032

Author(s):

Mihai Vasile Marinca ◽

Irina Draga Caruntu ◽

Ludmila Liliac ◽

Simona Eliza Giusca ◽

Andreea Marinca ◽

...

Keyword(s):

Sample Size ◽

Small Sample Size ◽

Germ Cell Tumors ◽

Disease Free Survival ◽

Developed Countries ◽

Small Sample ◽

P Value ◽

Testicular Germ Cell ◽

Therapeutic Decisions ◽

The Impact

e15032 Background: The 1997 IGCCCG Consensus classification provides clinicians with enough information to efficiently choose between treatment options for most GCT patients. Nevertheless, therapy is ineffective in 5-10% of cases (even more in less developed countries), and about the same numbers experience severe side effects. This exploratory study aims to assess the impact of more rigorous and detailed pathology examination on improving the assignation of these patients to prognostic groups and, consequently, making optimal therapeutic decisions. Methods: Predefined features were reviewed on histology slides from 39 GCT patients followed-up for a median of 48.28 months. We designed a uniform pathology protocol, focused on identifying potential new prognostic factors. Categorical and continuous variables were quantified using light microscopy and computer-aided morphometry and, due to the small sample size, their statistical correlation was analyzed by exact tests and Spearman’s rho, respectively. Significant (2-sided p-value <0.05, under sample size reserve) coefficient values were entered in hierarchical cluster analysis (HCA). Results: Favorable IGCCCG group, presence of seminoma, glandular tissue pattern, presence and histoarchitecture of lymphocytic infiltrate associated better survival rates and lower risk of progression. Invasion of the epididymis and spermatic cord, presence of teratoma, choriocarcinoma and yolk-sac elements, papillary pattern and cell pleomorphism predicted poorer outcomes. HCA yielded 2 significantly distinct patient groups in terms of overall survival (p=0.018) and time to progression (p=0.080), but not disease-free survival (p=0.614). Conclusions: Quantification of tumor subtypes and other histology features of GCTs (e.g. necrosis, tissue patterns, inflammation) is feasible and, if standardized, may prove useful in optimal selection of risk groups, when performed by an experienced pathologist.

Download Full-text

Seroprevalence of torch infections among complicated pregnancies in Atbara river Nile state - Sudan

Women s Health ◽

10.15406/mojwh.2021.10.00281 ◽

2021 ◽

Vol 10 (1) ◽

pp. 18-20

Author(s):

Mohammed Ahmed Ibrahim Ahmed ◽

Hanaa Elzain Musaad ◽

Nahla Ahmed Mohammed Abdurrahman ◽

Wadei Mohammed Yasin ◽

Mosab Nouraldein Mohammed Hamad ◽

...

Keyword(s):

Pregnant Women ◽

Small Sample Size ◽

Education Level ◽

Small Sample ◽

P Value ◽

River Nile ◽

Serological Evidence ◽

Chi Square ◽

Cross Sectional ◽

Torch Infection

Objective: The aim of this study was to find the seroprevalence of TORCH infection among volunteered pregnant women and to find out the correlation between TORCH infection and age, complicated pregnancies and education level. Methods: A descriptive cross-sectional laboratory-based study carried out between March to June 2012 at Atbara River Nile State northern Sudan. The statistical descriptive determined in means of percentages. Chi-square used for trend analysis and calculates p-values. Results: A total of (75) voluntary pregnant women, Age between 15-55 years. The most common age group was 26-35 years (37/75) 49%.Serological evidence of Toxoplasmosis were positive for (22/75) 29 %. All volunteered pregnant women were negative for other components of TORCH complex. Study showed non-significant correlation between age, education level and complicated pregnancies, P-value 0.13, 0.43 and 0.51 respectively. Conclusion: Low prevalence of Toxoplasmosis and absent of serological evidence of other TORCH complex among volunteered pregnant women in Atbara has been documented. Insignificant correlation between toxoplasmosis and age, education level as well as number of complicated pregnancies reported. This attributed, in part, to the small sample size and using screening tool as diagnostic modalities.

Download Full-text

Prevention of chemotherapy-induced delayed nausea and vomiting with aprepitant in patients recieving highly emetogenic-five day cisplatin regimens

Journal of Clinical Oncology ◽

10.1200/jco.2007.25.18_suppl.14125 ◽

2007 ◽

Vol 25 (18_suppl) ◽

pp. 14125-14125 ◽

Cited By ~ 1

Author(s):

A. J. Joshi ◽

H. Singh ◽

S. Chawla

Keyword(s):

Small Sample Size ◽

Rescue Therapy ◽

Small Sample ◽

P Value ◽

High Dose ◽

Delayed Emesis ◽

End Point ◽

Group I ◽

Delayed Nausea ◽

Group Ii

14125 Background: It has been shown that addition of the NK1 receptor antagonist aprepitant to 5HT3 antagonist plus dexamethasone is more effective than just the 5HT3 antagonist plus dexamethasone for prevention of acute and delayed emesis due to high dose cisplatin and also that this effect lasts for multiple cycles. This study evaluated whether the antiemetic efficacy of aprepitant could be sustained for 5 day cisplatin regimens. Methods: Patients receiving cisplatin 20mg/m2/day for 5 days (PEB and TCF regimens ) were randomized to one of the following two regimens: (1) aprepitant 125 mg 1 hour before cisplatin on day 1 and aprepitant 80 mg on days 2 and 3 (n = 17); (2) placebo before cisplatin on days 2 to 7 (n = 19). All groups received ondansetron 32 mg and dexamethasone 20 mg before cisplatin, and dexamethasone 8 mg on days 2 to 7. The primary end point was complete response (no emesis and no rescue therapy) over 7 days following cisplatin in up to six cycles analyzed by a modified intent-to-treat approach. Secondary end point was evaluated using Functional Living Index-Emesis questionnaire. Treatment comparisons were made using logistic regression models and P value calculated using the chi square test due to small sample size. Results: In the acute period, 83% and 56% of patients were without emesis in groups I and II, respectively (P < .01 for group I v group II). In the delayed period upto day 5, the proportion of patients without emesis in groups I and II, was 59% and 32%, respectively (P < .01 for groups I v group II). In the extended period day 6 and 7 the proportion of patients without emesis in groups I and II was 50% and 38% respectively (P< .01 for groups I v II). The distribution of nausea scores in the delayed period beyond day 5 was lower when comparing group I with group II (P < .05 for days 6 and 7). Two serious adverse events of diarrhea were probably attributed to aprepitant. Conclusions: Once daily oral administration of aprepitant was effective and superior in reducing delayed emesis and nausea after 5 days cisplatinum regimen when added to 5HT3 antagonist plus dexamethasone. This benefit persists upto day 7. Confirming and extending previous results aprepitant should be used in triple combination in patients receiving 5 day cisplatin regimens. No significant financial relationships to disclose.

Download Full-text

Neoadjuvant bevacizumab: Surgical complications of mastectomy with and without reconstruction.

Journal of Clinical Oncology ◽

10.1200/jco.2013.31.15_suppl.1100 ◽

2013 ◽

Vol 31 (15_suppl) ◽

pp. 1100-1100

Author(s):

Kari Joanne Kansal ◽

Laura Stewart Dominici ◽

Sara M. Tolaney ◽

Steven J. Isakoff ◽

Ian E. Krop ◽

...

Keyword(s):

Breast Cancer ◽

Neoadjuvant Therapy ◽

Surgical Complications ◽

Small Sample Size ◽

Small Sample ◽

P Value ◽

Data Sets ◽

Complication Rates ◽

Similar Time ◽

Time Period

1100 Background: Neoadjuvant therapy is commonly used in operable breast cancer. We prospectively evaluated the surgical complications in a cohort of patients who underwent mastectomy following neoadjuvant doxorubucin hydrochloride/cyclophosphamide/paclitaxel (AC/T) plus bevacizumab and compared the rate of complications to a matched cohort of neoadjuvant AC/T without bevacizumab. Methods: One hundred patients with HER2-negative breast cancer enrolled in a single-arm trial of neoadjuvant AC/T plus bevacizumab (cohort 1), 60 of these patients underwent mastectomy and were matched with 59 patients who received standard neoadjuvant AC/T (cohort 2) over a similar time period in the same healthcare system. All patients underwent mastectomy with or without reconstruction. Fisher’s exact tests were used to compare complication rates, with a p<0.05 was considered significant. Results: Patients were matched well in terms of demographics. The overall complication rate was 33% in cohort 1 and 31% in cohort 2 (P-value=0.84; Table). In cohort 1, 7 of 23 (30%) patients who underwent immediate expander/implant reconstruction had complications, including 2 patients who had explantation of their reconstructions. In cohort 2, 0 of 8 (0%) had complications (p value=0.15). Conclusions: Nearly a third of patients undergoing neoadjuvant therapy with AC/T with or without bevacizumab developed a postoperative complication after mastectomy. The use of bevacizumab was not associated with a significant increase in surgical complications, although this is a non-randomized data with a small sample size. As larger data sets become available with the use of neoadjuvant bevacizumab with mastectomy, further refinement may be necessary. [Table: see text]

Download Full-text

Abstract P113: Lipid-related Genetic Variants and Lipid Outcomes in a Cohort of Chilean Children

Circulation ◽

10.1161/circ.135.suppl_1.p113 ◽

2017 ◽

Vol 135 (suppl_1) ◽

Author(s):

Ann Von Holle ◽

Anne Justice ◽

Kari E North ◽

Bárbara Angel ◽

Estela Blanco ◽

...

Keyword(s):

Genetic Variants ◽

Genetic Model ◽

Small Sample Size ◽

A Priori ◽

Density Lipoprotein ◽

The United States ◽

Small Sample ◽

Lipoprotein Cholesterol ◽

Risk Scores ◽

Lipid Levels

Dyslipidemia is an important risk factor for chronic cardiometabolic diseases. Lipid traits are highly heritable and there are currently >185 established loci influencing lipid levels in adults. Recent studies have confirmed that variants associated with lipids influence lipid levels across the lifecourse, and in ancestrally diverse populations. Given that Hispanic/Latinos (HL) shoulder much of the cardiometabolic burden in the United States, it is important to identify genetic variants that contribute the greatest risk for elevated lipid levels across life stages. Thus, our primary aim is to examine the association of known lipid variants with lipid traits identified in large study of adult participants from a Chilean infancy cohort of primarily European-descent. The sample assessed from 2008 to 2013 (n=546) had genotyping and well-measured lipid phenotypes (median age: 16.8 years, interquartile range: 16.6, 16.9). We assessed single variant associations using linear regression for high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C) and triglycerides (TG), assuming an additive genetic model, adjusted for sex. Additionally, we regressed phenotypes onto weighted trait-specific polygenic risk scores (PRS). Only six variants from the Chilean sample met the a priori threshold of power > 0.8. We found statistically significant effect sizes (mmol/l (se)) for four of the six variants: rs3764261 (0.16 (0.04)) and rs1532085 (0.05 (0.04)) for HDL and rs1260326 (0.34 (0.15)) and rs964184 (0.33 (0.15)) for TG. For each significant variant, direction of effect matched the multiethnic adult GWAS from which SNPs were selected. We compared our findings to a previous study in Finnish children at age 18 years (n=1,216) and found an opposite direction of effect for our significant HDL variants. Likewise, when comparing coefficients for the PRS between the Chilean and Finnish youth sample we found the association to be stronger in the Chilean sample for every trait and gender group with the exception of LDL for males. The lipid loci explained the least amount of total variance for LDL (males=4% and females=5%) and the most amount of variance for HDL (males=20% and females=14%). In conclusion, there is evidence that lipid loci from a HL sample of adolescents contain similar associations as those from European children and adults. Despite the small sample size and possibility for bias with different ancestral groups we found meaningful and statistically significant associations relating lipid loci in a HL cohort of Chilean adolescents with those found in European ancestral groups. These associations emphasize the importance of adolescence as a time for disease prevention given studies demonstrating both the persistence of associations between PRS and lipids over the life course and the increasing role PRS plays in predicting disease.

Download Full-text

Regression shrinkage methods for clinical prediction models do not guarantee improved performance: Simulation study

Statistical Methods in Medical Research ◽

10.1177/0962280220921415 ◽

2020 ◽

Vol 29 (11) ◽

pp. 3166-3178 ◽

Cited By ~ 2

Author(s):

Ben Van Calster ◽

Maarten van Smeden ◽

Bavo De Cock ◽

Ewout W Steyerberg

Keyword(s):

Maximum Likelihood ◽

Sample Size ◽

Simulation Study ◽

Prediction Models ◽

Small Sample Size ◽

Predictive Performance ◽

Small Sample ◽

Adaptive Lasso ◽

Shrinkage Methods ◽

Improved Performance

When developing risk prediction models on datasets with limited sample size, shrinkage methods are recommended. Earlier studies showed that shrinkage results in better predictive performance on average. This simulation study aimed to investigate the variability of regression shrinkage on predictive performance for a binary outcome. We compared standard maximum likelihood with the following shrinkage methods: uniform shrinkage (likelihood-based and bootstrap-based), penalized maximum likelihood (ridge) methods, LASSO logistic regression, adaptive LASSO, and Firth’s correction. In the simulation study, we varied the number of predictors and their strength, the correlation between predictors, the event rate of the outcome, and the events per variable. In terms of results, we focused on the calibration slope. The slope indicates whether risk predictions are too extreme (slope < 1) or not extreme enough (slope > 1). The results can be summarized into three main findings. First, shrinkage improved calibration slopes on average. Second, the between-sample variability of calibration slopes was often increased relative to maximum likelihood. In contrast to other shrinkage approaches, Firth’s correction had a small shrinkage effect but showed low variability. Third, the correlation between the estimated shrinkage and the optimal shrinkage to remove overfitting was typically negative, with Firth’s correction as the exception. We conclude that, despite improved performance on average, shrinkage often worked poorly in individual datasets, in particular when it was most needed. The results imply that shrinkage methods do not solve problems associated with small sample size or low number of events per variable.

Download Full-text