Modeling Variables With a Spike at Zero: Examples and Practical Recommendations

Opioid and non-opioid analgesic regimens after fracture and risk of serious opioid-related events

Background

Non-opioid analgesics are prescribed in combination with opioids among patients with long bone fracture to reduce opioid prescribing needs, yet evidence is limited on whether they reduce the risk of serious opioid-related events (SOREs). We compared the risk of SOREs among hospitalized patients with long bone fracture discharged with filled opioid prescriptions, with and without non-opioid analgesics.

Design

We identified a retrospective cohort of analgesic-naïve adult patients with a long bone fracture hospitalization using the Merative MarketScan Commercial Database (2013-2020). The exposure was opioid and non-opioid analgesic (gabapentinoids, muscle relaxants, non-steroidal anti-inflammatory drugs, acetaminophen) prescriptions filled in the 3 days before through 42 days after discharge. The outcome was the development of new persistent opioid use or opioid use disorder during follow-up (day 43 through day 408 after discharge). We used Cox proportional hazards regression with inverse probability of treatment weighting with overlap trimming to compare outcomes among those that filled an opioid and a non-opioid analgesic to those that filled only an opioid analgesic. In secondary analyses, we used separate models to compare those that filled a prescription for each specific non-opioid analgesic type with opioids to those that filled only opioids.

Results

Of 29 489 patients, most filled an opioid prescription alone (58.4%) or an opioid and non-opioid (22.0%). In the weighted proportional hazards regression model accounting for relevant covariates and total MME, filling both a non-opioid analgesic and an opioid analgesic was associated with 1.63 times increased risk of SOREs compared with filling an opioid analgesic only (95% CI 1.41 to 1.89). Filling a gabapentin prescription in combination with an opioid was associated with an increased risk of SOREs compared with those that filled an opioid only (adjusted HR: 1.84 (95% CI1.48 to 2.27)).

Conclusions

Filling a non-opioid analgesic in combination with an opioid was associated with an increased risk of SOREs after long bone fracture.

Level of evidence

Level III, prognostic/epidemiological.

Study type

Retrospective cohort study.

Prescription Opioid Exposure During Pregnancy and Risk of Spontaneous Preterm Delivery

Importance

Opioid exposure during pregnancy has been associated with preterm birth, but prior studies have not differentiated between spontaneous and indicated preterm birth or fully investigated these associations as functions of opioid dose.

Objective

To determine whether prescription opioid use during pregnancy is associated with spontaneous preterm birth and whether the association is dose-dependent.

Design, Setting, and Participants

This case-control study examined a retrospective cohort of pregnant patients enrolled in Tennessee Medicaid. Enrollment files were linked to health care encounters, hospital discharge information, birth certificate data, and prescription fills. Eligible participants were pregnant people ages 15 to 44 years without opioid use disorder who experienced birth of a single fetus at 24 weeks gestation or greater between 2007 and 2019 with linked birth certificate data. Cases of spontaneous preterm birth were matched with up to 10 controls based on pregnancy start date, race, ethnicity, age at delivery within 2 years, and history of prior preterm birth. Cases and matched controls were continuously enrolled in TennCare for at least 90 days prior to the index date (case delivery date).

Exposure

Total opioid MME filled during the 60 days prior to the index date.

Main Outcomes and Measures

The primary outcome was spontaneous preterm birth determined by a validated algorithm using birth certificate data. Conditional logistic regression was used to estimate the association between spontaneous preterm birth and total opioid morphine milligram equivalents (MME) dispensed, adjusting for parity, prepregnancy body mass index, education level, tobacco use, hepatitis infections, and pain indications.

Results

A total of 25 391 cases (median [IQR] age, 23 [20-28] years; 127 Asian [0.5%], 9820 Black [38.7%], 664 Hispanic [2.6%]; 14 748 non-Hispanic White [58.1%]) with spontaneous preterm birth were identified and matched with 225 696 controls (median [IQR] age, 23 [20-27] years; 229 Asian [0.1%], 89 819 Black [39.8%], 3590 Hispanic [1.6%]; 132 002 non-Hispanic White [58.5%]) (251 087 patients total), with 18 702 patients (7.4%) filling an opioid prescription in the 60 days prior to the index date. Each doubling of nonzero opioid MME was associated with a 4% increase in the odds of spontaneous preterm birth compared with no opioid exposure (adjusted odds ratio, 1.04; 95% CI, 1.01-1.08).

Conclusions and Relevance

In this case-control study, a positive association was found between total prescription opioid dose dispensed and the odds of spontaneous preterm birth. These findings support guidance to minimize opioid exposure during pregnancy and prescribe the lowest dose necessary.

Sample size and predictive performance of machine learning methods with survival data: A simulation study

Prediction models are increasingly developed and used in diagnostic and prognostic studies, where the use of machine learning (ML) methods is becoming more and more popular over traditional regression techniques. For survival outcomes the Cox proportional hazards model is generally used and it has been proven to achieve good prediction performances with few strong covariates. The possibility to improve the model performance by including nonlinearities, covariate interactions and time-varying effects while controlling for overfitting must be carefully considered during the model building phase. On the other hand, ML techniques are able to learn complexities from data at the cost of hyper-parameter tuning and interpretability. One aspect of special interest is the sample size needed for developing a survival prediction model. While there is guidance when using traditional statistical models, the same does not apply when using ML techniques. This work develops a time-to-event simulation framework to evaluate performances of Cox regression compared, among others, to tuned random survival forest, gradient boosting, and neural networks at varying sample sizes. Simulations were based on replications of subjects from publicly available databases, where event times were simulated according to a Cox model with nonlinearities on continuous variables and time-varying effects and on the SEER registry data.

Alcohol intake and the risk of epithelial ovarian cancer

PURPOSE

To investigate the association between alcohol intake over the lifetime and the risk of overall, borderline, and invasive ovarian cancer.

METHODS

In a population-based case-control study of 495 cases and 902 controls, conducted in Montreal, Canada, average alcohol intake over the lifetime and during specific age periods were computed from a detailed assessment of the intake of beer, red wine, white wine and spirits. Multivariable logistic regression was used to estimate odds ratios (ORs) with 95% confidence intervals (CIs) for the association between alcohol intake and ovarian cancer risk.

RESULTS

For each one drink/week increment in average alcohol intake over the lifetime, the adjusted OR (95% CI) was 1.06 (1.01-1.10) for ovarian cancer overall, 1.13 (1.06-1.20) for borderline ovarian cancers and 1.02 (0.97-1.08) for invasive ovarian cancers. This pattern of association was similarly observed for alcohol intake in early (15- < 25 years), mid (25- < 40 years) and late adulthood (≥ 40 years), as well as for the intake of specific alcohol beverages over the lifetime.

CONCLUSIONS

Our results support the hypothesis that a higher alcohol intake modestly increases the risk of overall ovarian cancer, and more specifically, borderline tumours.

Iron Status and Associated Factors among Canadian Women: Results from the Canadian Health Measures Survey

BACKGROUND

Iron deficiency and overload may negatively impact women's health. There has been limited assessment of iron status and its associated factors among Canadian women.

OBJECTIVES

This study investigated associations of various sociodemographic, lifestyle, medication, and dietary factors with body iron stores among pre- and postmenopausal women in Canada.

METHODS

Analyses were conducted using cross-sectional, nationally representative survey and biomarker data from women aged 20-79 y (n = 6362) in the Canadian Health Measures Survey (2009-2017). Body iron stores were assessed by measuring serum concentrations of ferritin (SF). Information on potential correlates was collected during an in-home interview. Multivariable linear regression analyses were performed to evaluate associations with SF concentration, and logistic regression was used to estimate associations with iron deficiency (SF <15 μg/L) or elevated iron stores (SF >150 μg/L).

RESULTS

Geometric mean SF concentrations were significantly higher in postmenopausal than in premenopausal women (73.2 versus 33.8 μg/L; P < 0.001). The prevalence of iron deficiency among pre- and postmenopausal women was 16.0% and 4.0%, respectively, whereas that of elevated iron stores was 2.7% and 21.0%, respectively. After simultaneous adjustment for multiple factors, including high-sensitivity CRP (inflammation marker), we found that age, East/Southeast Asian (versus White) race/ethnicity, alcohol, and red meat consumption were positively associated with SF concentration among pre- and postmenopausal women. In addition, aspirin use and dairy consumption were inversely associated with SF concentration among postmenopausal women only. Similar patterns were observed for associations with elevated iron stores among postmenopausal women, whereas higher grain consumption was associated with an increased prevalence of iron deficiency among premenopausal women.

CONCLUSIONS

Sociodemographic, lifestyle, medication, and dietary factors are correlated with iron status determined by SF concentration among Canadian women. The findings may have implications for intervention strategies aimed at optimizing body iron stores in pre- and postmenopausal women.

Long-term exposure to wildfires and cancer incidence in Canada: a population-based observational cohort study

BACKGROUND

Wildfires emit many carcinogenic pollutants that contaminate air, water, terrestrial, and indoor environments. However, little is known about the relationship between exposure to wildfires and cancer risk. We aimed to assess the associations between residential exposure to wildfires and the incidence of several cancer outcomes (lung cancer, brain cancer, non-Hodgkin lymphoma, multiple myeloma, and leukaemia) in Canada.

METHODS

We did a population-based observational cohort study of participants in the 1996 Canadian Census Health and Environment Cohort. The 1996 Canadian Census Health and Environment Cohort is a nationally representative sample of Canadian adults, followed up for cancer incidence and mortality from 1996 to 2015. For this analysis, we excluded participants who lived in major Canadian cities (with a population size greater than 1·5 million people), recent immigrants, and individuals younger than 25 years or 90 years of age or older at baseline. Exposures to wildfires were assigned on the basis of area burned within a 20 km or 50 km radius of residential locations and updated for annual residential mobility. Multivariable Cox proportional hazards models were used to estimate associations between exposure to wildfires and specific cancers associated with carcinogenic compounds released by wildfires, including lung and brain cancer, non-Hodgkin lymphoma, multiple myeloma, and leukaemia, adjusted for many personal and neighbourhood-level covariates.

FINDINGS

Our analyses included more than 2 million people followed up for a median of 20 years, for a total of 34 million person-years. Wildfire exposure was associated with slightly increased incidence of lung cancer and brain tumours. For example, cohort members exposed to a wildfire within 50 km of residential locations in the past 10 years had a 4·9% relatively higher incidence (adjusted hazard ratio [HR] 1·049, 95% CI 1·028-1·071) of lung cancer than unexposed populations, and a 10% relatively higher incidence (adjusted HR 1·100, 1·026-1·179) of brain tumours. Similar associations were observed for the 20 km buffer size. Wildfires were not associated with haematological cancers in this study, and concentration-response trends were not readily apparent when area burned was modelled as a continuous variable.

INTERPRETATION

Long-term exposure to wildfires might increase the risk of lung cancer and brain tumours. Further work is needed to develop long-term estimates of wildfire exposures that capture the complex mixture of environmental pollutants released during these events.

FUNDING

Canadian Institute for Health Research and Fonds de recherche du Quebec.

Statistical Analysis in the German National Cohort (NAKO) - Specific Aspects and General Recommendations

The German National Cohort (NAKO) is an ongoing, prospective multicenter cohort study, which started recruitment in 2014 and includes more than 205,000 women and men aged 19-74 years. The study data will be available to the global research community for analyses. Although the ultimate decision about the analytic methods will be made by the respective investigator, in this paper we provide the basis for a harmonized approach to the statistical analyses in the NAKO. We discuss specific aspects of the study (e.g., data collection, weighting to account for the sampling design), but also give general recommendations which may apply to other large cohort studies as well.

The Expression of Programmed Death Ligand 1 and Vimentin in Resected Non-Metastatic Non-Small-Cell Lung Cancer: Interplay and Prognostic Effects

Programmed death ligand 1 (PD-L1) is an immune checkpoint with a role in cancer-related immune evasion. It is a target for cancer immunotherapy and its expression is detected for the use of some immune checkpoint inhibitors in advanced non-small cell lung cancer patients (NSCLC). Vimentin is a key component of the epithelial-to-mesenchymal transition phenotype. Its expression has negative prognostic effects in NSCLC. In this study, we retrospectively evaluated PD-L1 and vimentin expression in tumor cells, immune infiltrate and PD-L1 positive immune infiltrate via immunohistochemistry in tissue samples from resected non-metastatic NSCLC patients. We explored the interplay between PD-L1 and vimentin expression through Spearman’s correlation test. We performed univariate analysis through the Cox models for demographic and clinico-pathological variables, and also for dichotomized biomarkers, i.e., PD-L1 and vimentin in tumor cells, both with 1 and 50% cutoffs. We used Kaplan-Meier method to estimate the overall survival, comparing both vimentin and PD-L1 positive patients with all the others. We found a weak positive correlation between PD-L1 and vimentin expressions in tumor cells (r = 0.25; p = 0.001). We also observed a statistically not significant trend towards a shorter overall survival in patients with both PD-L1 and vimentin expression >1% (HR = 1.36; 95% CI: 0.96–1.93, p = 0.087). In conclusion, these findings suggest that interplay between PD-L1 and vimentin may exist in non-metastatic NSCLC patients and the positivity of both markers in tumor tissue is associated with a trend towards a worse prognosis.

Newly Identified Gross Human Anatomy: Eight Paired Vestigial Breast Mounds Run along the Embryological Mammary Ridges in Lean Adults

Although isolated accessory breasts have been reported in many locations on the torso, we noted that lean men presenting for liposuction had mounds of fullness in a curvilinear array (axilla to groin) consistent with the configuration of the embryological mammary ridge. We hypothesized that grid mapping and pinch testing would elucidate the pattern of fullness and its relation to the "milk line."

METHODS

Twenty lean participants (10 men, 10 women) each underwent standardized photography and grid mapping of anterior torsos. We then pinch-tested each for subcutaneous fullness in 250 standardized loci. We used plotted pinch values and OLS regression models to determine if focal fullness corresponded to published configurations of the embryologic mammary ridge.

RESULTS

We identified a pattern of paired mounds in all participants running bilaterally in curved linear arrays from axillae to groin in the exact form of milk lines. Regression models applied to the male and female pinch data indicated that focal thickness matched mammary ridge configurations in all subjects (P = 0.023 for men, <0.001 for women). Fatty fullness never appeared elsewhere on the anterior torsos.

CONCLUSIONS

The linear pattern of paired focal fat mounds present on the anterior torsos is consistent with the paths of embryological mammary ridges, suggesting these focal fat pads are of breast origin. The consistent mounds are distinct from general subcutaneous fullness and therefore represent a new finding of gross human anatomy. We recommend further research to define differential physiology of distinct subpopulations of yellow fat and clinical implications.

The risk of serious opioid-related events associated with common opioid prescribing regimens in the postpartum period after cesarean delivery

BACKGROUND

Opioid analgesics are commonly prescribed to women after cesarean delivery. There is a growing effort to prescribe opioids judiciously; however, the risk of serious opioid-related events associated with specific prescribing patterns after cesarean delivery remains unclear.

OBJECTIVE

We examined the association between the dosage of the first opioid prescription filled after cesarean delivery and the risk of serious opioid-related events.

STUDY DESIGN

We identified opioid-naïve women with a cesarean delivery enrolled in Tennessee Medicaid (2007-2014). Pharmacy prescription fill data characterized opioids filled within 5 days after delivery. Patients were followed up from day 5 after delivery to the earliest of the following: serious opioid-related event (persistent opioid use, evidence of opioid use disorder [diagnosis or methadone or buprenorphine fill], overdose, or opioid-related death), non-opioid-related death, enrollment loss, or 365th day. We estimated the adjusted hazard ratios and 95% confidence intervals for the serious opioid-related event outcomes based on the dosage (morphine milligram equivalents) of the first filled opioid prescription, adjusting for baseline sociodemographic characteristics, delivery complications, multiple deliveries, comorbidities, and medication use. Secondary analyses examined the role of commonly prescribed opioid strengths and quantities.

RESULTS

The overall incidence rate of serious opioid-related events among women after cesarean delivery was 3.0 per 100 person-years. Compared with women who did not fill an opioid prescription, the rate of serious opioid-related events was higher among women who filled an opioid prescription, although only significantly higher among women who filled a total dosage of ≥100 morphine milligram equivalents (97.1% of opioid prescriptions). In the secondary analyses, women with a low prescribed daily opioid dosage and women with a low prescribed number of oxycodone (5 mg) tablets (<10 tablets) were not at increased risk of serious opioid-related events compared with women who did not fill an opioid prescription.

CONCLUSION

Opioid-naïve women who filled a postpartum opioid prescription at commonly prescribed doses after cesarean delivery had an increased risk of serious opioid-related events compared to women who did not fill a postpartum opioid prescription. Low opioid doses were not associated with a significant increase in the risk of serious opioid-related events.

Association between physical activity, grip strength and sedentary behaviour with incidence of malignant melanoma: results from the UK Biobank

BACKGROUND

Physical activity has been positively related to malignant melanoma. However, that association may be confounded by ultraviolet radiation (UV), a variable closely related to both outdoor physical activity and malignant melanoma. We examined physical activity, grip strength and sedentary behaviour in relation to risk of malignant melanoma, accounting for relevant confounders using data from a prospective cohort study.

METHODS

In 350,512 UK Biobank participants aged 38-73 years at baseline, physical activity was assessed with a modified version of the International Physical Activity Questionnaire Short Form, grip strength was measured with a hand dynamometer, and sedentary behaviour was recorded with three specific questions. Multivariable hazard ratios (HR) and corresponding 95% confidence intervals (CI) were estimated using Cox proportional hazards regression.

RESULTS

During 7 years of follow-up, 1239 incident malignant melanoma diagnoses were recorded. Physical activity and sedentary behaviour were unrelated to malignant melanoma (HRs 1.01 (95% CI 0.95-1.07) and 1.04 (95% CI 0.97-1.12), respectively), and the initially positive association with grip strength in the basic model (HR 1.23, 95% CI 1.08-1.40) was attenuated after full adjustment (HR 1.10, 95% CI 0.96-1.26).

CONCLUSION

Physical activity, grip strength and sedentary behaviour are not associated with malignant melanoma risk.

Caffeinated Coffee Consumption and Health Outcomes in the US Population: A Dose-Response Meta-Analysis and Estimation of Disease Cases and Deaths Avoided

To explore the role of coffee on health outcomes in the United States, where coffee consumption is common, we conducted a meta-analysis of prospective studies investigating the magnitude (any compared with no consumption) and the dose-response shape (cups per day) of the associations between caffeinated coffee consumption and incidence/mortality of cardiovascular disease (CVD), as well as incidence of type 2 diabetes (T2D), hepatocellular carcinoma (HCC), endometrial cancer, melanoma, and nonmelanoma skin cancer. We selected the desirable health outcomes that have been shown to be positively associated with coffee consumption. Studies were identified by searching PubMed/Embase databases up to September 2019. Inclusion criteria included prospective studies that investigated the relation of ≥3 categories of caffeinated coffee consumption and the outcomes of interest. Twenty-six studies (42 distinct cohorts), with 93,706 cases/deaths and 3,713,932 participants, met the inclusion criteria. In any coffee consumers, there was a significant inverse association with the risk of CVD (RR = 0.90; 95% CI: 0.84, 0.96), T2D (RR = 0.90; 95% CI: 0.85, 0.96), endometrial cancer (RR = 0.85; 95% CI: 0.78, 0.92), melanoma (RR = 0.89; 95% CI: 0.80, 0.99), and nonmelanoma skin cancer (RR = 0.92; 95% CI: 0.89, 0.95). Coffee consumption was also inversely associated with HCC (RR = 0.93; 95% CI: 0.80, 1.08), without reaching statistical significance. The dose-response relation was nonlinear uniquely for CVD (P-nonlinearity = 0.01). In particular, the largest risk reduction was observed for 3-4 cups/d (∼120 mL/cup) and no reduction thereafter. For other outcomes, the risk decreased linearly over the whole coffee consumption range. Current patterns of consumption in the United States would account for a fraction of avoided cases/deaths ranging from 6% to 12% according to the outcome considered. This study confirms the beneficial health effects of caffeinated coffee consumption in the US population on the health outcomes considered, and quantifies their possible magnitude.

Models to Assess the Association of a Semiquantitative Exposure With Outcomes

A semiquantitative risk factor has 2 components: any exposure (yes/no) and the quantitative amount of exposure (if exposed). We describe the statistical properties of alternative analyses with such a risk factor using linear, logistic, or Cox proportional hazards models. Often analyses employ the amount exposed as a single quantitative covariate, including the nonexposed with value zero. However, this analysis provides a biased estimate of the exposure coefficient (slope) and we describe the magnitude of the bias. This bias can be eliminated by adding a binary covariate for exposed versus not to the model. This 2-factor analysis captures the full risk-factor effect on the outcome. However, the coefficient for any exposure versus not does not have a meaningful interpretation. Alternatively, when exposure values among those exposed are centered (by subtracting the mean), the estimate of this coefficient represents the difference in the outcome between those exposed versus not in aggregate. We also show that the biased model provides biased estimates of the coefficients for other covariates added to the model. Proper analysis of a semiquantitative risk factor should start with a 2-factor model, with centering, to assess the joint contributions of the 2 components of the risk-factor exposure. Properties of models were illustrated using data from a multisite study in North America (1983-2019).

Short-term effects of rainfall on childhood hand, foot and mouth disease and related spatial heterogeneity: evidence from 143 cities in mainland China

BACKGROUND

Numerous studies have demonstrated the potential association between rainfall and hand, foot and mouth disease (HFMD), but the results are inconsistent. This study aimed to quantify the relationship between rainfall and HFMD based on a multicity study and explore the potential sources of spatial heterogeneity.

METHODS

We retrieved the daily counts of childhood HFMD and the meteorological variables of the 143 cities in mainland China between 2009 and 2014. A common time series regression model was applied to quantify the association between rainfall and HFMD for each of the 143 cities. Then, we adopted the meta-regression model to pool the city-specific estimates and explore the sources of heterogeneity by incorporating city-specific characteristics.

RESULTS

The overall pooled estimation suggested a nonlinear exposure-response relationship between rainfall and HFMD. Once rainfall exceeded 15 mm, the HFMD risk stopped increasing linearly and began to plateau with the excessive risk ratio (ERR) peaking at 21 mm of rainfall (ERR = 3.46, 95% CI: 2.05, 4.88). We also found significant heterogeneity in the rainfall-HFMD relationships (I² = 52.75%, P < 0.001). By incorporating the city-specific characteristics into the meta-regression model, temperature and student density can explain a substantial proportion of spatial heterogeneity with I² statistics that decreased by 5.29 and 6.80% at most, respectively.

CONCLUSIONS

Our findings verified the nonlinear association between rainfall and HFMD. The rainfall-HFMD relationship also varies depending on locations. Therefore, the estimation of the rain-HFMD relationship of one location should not be generalized to another location.

State of the art in selection of variables and functional forms in multivariable analysis-outstanding issues

BACKGROUND

How to select variables and identify functional forms for continuous variables is a key concern when creating a multivariable model. Ad hoc 'traditional' approaches to variable selection have been in use for at least 50 years. Similarly, methods for determining functional forms for continuous variables were first suggested many years ago. More recently, many alternative approaches to address these two challenges have been proposed, but knowledge of their properties and meaningful comparisons between them are scarce. To define a state of the art and to provide evidence-supported guidance to researchers who have only a basic level of statistical knowledge, many outstanding issues in multivariable modelling remain. Our main aims are to identify and illustrate such gaps in the literature and present them at a moderate technical level to the wide community of practitioners, researchers and students of statistics.

METHODS

We briefly discuss general issues in building descriptive regression models, strategies for variable selection, different ways of choosing functional forms for continuous variables and methods for combining the selection of variables and functions. We discuss two examples, taken from the medical literature, to illustrate problems in the practice of modelling.

RESULTS

Our overview revealed that there is not yet enough evidence on which to base recommendations for the selection of variables and functional forms in multivariable analysis. Such evidence may come from comparisons between alternative methods. In particular, we highlight seven important topics that require further investigation and make suggestions for the direction of further research.

CONCLUSIONS

Selection of variables and of functional forms are important topics in multivariable analysis. To define a state of the art and to provide evidence-supported guidance to researchers who have only a basic level of statistical knowledge, further comparative research is required.

Detection of suspicious interactions of spiking covariates in methylation data

BACKGROUND

In methylation analyses like epigenome-wide association studies, a high amount of biomarkers is tested for an association between the measured continuous outcome and different covariates. In the case of a continuous covariate like smoking pack years (SPY), a measure of lifetime exposure to tobacco toxins, a spike at zero can occur. Hence, all non-smokers are generating a peak at zero, while the smoking patients are distributed over the other SPY values. Additionally, the spike might also occur on the right side of the covariate distribution, if a category "heavy smoker" is designed. Here, we will focus on methylation data with a spike at the left or the right of the distribution of a continuous covariate. After the methylation data is generated, analysis is usually performed by preprocessing, quality control, and determination of differentially methylated sites, often performed in pipeline fashion. Hence, the data is processed in a string of methods, which are available in one software package. The pipelines can distinguish between categorical covariates, i.e. for group comparisons or continuous covariates, i.e. for linear regression. The differential methylation analysis is often done internally by a linear regression without checking its inherent assumptions. A spike in the continuous covariate is ignored and can cause biased results.

RESULTS

We have reanalysed five data sets, four freely available from ArrayExpress, including methylation data and smoking habits reported by smoking pack years. Therefore, we generated an algorithm to check for the occurrences of suspicious interactions between the values associated with the spike position and the non-spike positions of the covariate. Our algorithm helps to decide if a suspicious interaction can be found and further investigations should be carried out. This is mostly important, because the information on the differentially methylated sites will be used for post-hoc analyses like pathway analyses.

CONCLUSIONS

We help to check for the validation of the linear regression assumptions in a methylation analysis pipeline. These assumptions should also be considered for machine learning approaches. In addition, we are able to detect outliers in the continuous covariate. Therefore, more statistical robust results should be produced in methylation analysis using our algorithm as a preprocessing step.

Elevated levels of eEF1A2 protein expression in triple negative breast cancer relate with poor prognosis

Eukaryotic elongation factor 1 alpha 2 (eEF1A2) is a translation factor selectively expressed by heart, skeletal muscle, nervous system and some specialized cells. Its ectopic expression relates with tumorigenesis in several types of human cancer. No data are available about the role of eEF1A2 in Triple Negative Breast Cancers (TNBC). This study investigated the relation between eEF1A2 protein levels and the prognosis of TNBC. A total of 84 TNBC diagnosed in the period 2002-2011 were included in the study. eEF1A2 protein level was measured in formalin-fixed paraffin-embedded tissues by immunohistochemistry in a semi-quantitative manner (sum of the percentage of positive cells x staining intensity) on a scale from 0 to 300. Cox regression assessed the association between eEF1A2 levels and disease-free survival (DFS) and breast cancer-specific survival (BCSS). Elevated values of eEF1A2 were associated with older age at diagnosis (p = 0.003), and androgen receptors positivity (p = 0.002). At univariate Cox analysis, eEF1A2 levels were not significantly associated with DFS and BCSS (p = 0.11 and p = 0.08, respectively) whereas adjusting for stage of disease, elevated levels of eEF1A2 protein resulted associated with poor prognosis (HR = 1.05, 95% CI: 1.01-1.11, p = 0.04 and HR = 1.07, 95% CI: 1.01-1.14, p = 0.03 for DFS and BCSS, respectively). This trend was confirmed analyzing negative versus positive samples by using categorized scores. Our data showed a negative prognostic role of eEF1A2 protein in TNBC, sustaining further investigations to confirm this result by wider and independent cohorts of patients.