Comparing measures of model selection for penalized splines in Cox models

Potential causal links and mediation pathway between urban greenness and lung cancer mortality: Result from a large cohort (2009 to 2020)

Urban greenness, as a vital component of the urban environment, plays a critical role in mitigating the adverse effects of rapid urbanization and supporting urban sustainability. However, the causal links between urban greenness and lung cancer mortality and its potential causal pathway remain poorly understood. Based on a prospective community-based cohort with 581,785 adult participants in southern China, we applied a doubly robust Cox proportional hazard model to estimate the causal associations between urban greenness exposure and lung cancer mortality. A general multiple mediation analysis method was utilized to further assess the potential mediating roles of various factors including particulate matter (PM1, PM2.5-1, and PM10-2.5), temperature, physical activity, and body mass index (BMI). We observed that each interquartile range (IQR: 0.06) increment in greenness exposure was inversely associated with lung cancer mortality, with a hazard ratio (HR) of 0.89 (95 % CI: 0.83, 0.96). The relationship between greenness and lung cancer mortality might be partially mediated by particulate matter, temperature, and physical activity, yielding a total indirect effect of 0.826 (95 % CI: 0.769, 0.887) for each IQR increase in greenness exposure. Notably, the protective effect of greenness against lung cancer mortality could be achieved primarily by reducing the particulate matter concentration.

Model design for nonparametric phylodynamic inference and applications to pathogen surveillance

Inference of effective population size from genomic data can provide unique information about demographic history and, when applied to pathogen genetic data, can also provide insights into epidemiological dynamics. The combination of nonparametric models for population dynamics with molecular clock models which relate genetic data to time has enabled phylodynamic inference based on large sets of time-stamped genetic sequence data. The methodology for nonparametric inference of effective population size is well-developed in the Bayesian setting, but here we develop a frequentist approach based on nonparametric latent process models of population size dynamics. We appeal to statistical principles based on out-of-sample prediction accuracy in order to optimize parameters that control shape and smoothness of the population size over time. Our methodology is implemented in a new R package entitled mlesky. We demonstrate the flexibility and speed of this approach in a series of simulation experiments and apply the methodology to a dataset of HIV-1 in the USA. We also estimate the impact of non-pharmaceutical interventions for COVID-19 in England using thousands of SARS-CoV-2 sequences. By incorporating a measure of the strength of these interventions over time within the phylodynamic model, we estimate the impact of the first national lockdown in the UK on the epidemic reproduction number.

A prediction model of cognitive impairment risk in elderly illiterate Chinese women

Objective

To establish and validate a targeted model for the prediction of cognitive impairment in elderly illiterate Chinese women.

Methods

1864 participants in the 2011-2014 cohort and 1,060 participants in the 2014-2018 cohort from the Chinese Longitudinal Healthy Longevity Survey (CLHLS) were included in this study. The Chinese version of the Mini-Mental State Examination (MMSE) was used to measure cognitive function. Demographics and lifestyle information were collected to construct a risk prediction model by a restricted cubic spline Cox regression. The discrimination and accuracy of the model were assessed by the area under the curve (AUC) and the concordance index, respectively.

Results

A total of seven critical variables were included in the final prediction model for cognitive impairment risk, including age, MMSE score, waist-to-height ratio (WHtR), psychological score, activities of daily living (ADL), instrumental abilities of daily living (IADL), and frequency of tooth brushing. The internal and external validation AUCs were 0.8 and 0.74, respectively; and the receiver operating characteristic (ROC) curves indicated good performance ability of the constructed model.

Conclusion

A feasible model to explore the factors influencing cognitive impairment in elderly illiterate women in China and to identify the elders at high risk was successfully constructed.

Long-term effect of intermediate particulate matter (PM1-2.5) on incident asthma among middle-aged and elderly adults: A national population-based longitudinal study

BACKGROUND

There is insufficient evidence about the long-term effects of intermediate particulate matter (PM_1-2.5) on asthma development in adults aged 45 years and above. This study aimed to investigate the relationship between long-term exposure to PM_1-2.5 and the incidence of asthma in adults aged 45 years and above.

METHODS

A cohort study based on the China Health and Retirement Longitudinal Study (CHARLS) database was conducted to investigate the long-term effects of PM_1-2.5 on self-reported asthma incidence in adults aged 45 years and above in China from 2011 to 2018. The PM concentrations were estimated using a high-resolution (1 km²) satellite-based spatiotemporal model. A covariate-adjusted generalized linear mixed model was used to analyze the relationship between long-term exposure to PM_1-2.5 and the incidence of asthma. Effect modifications and sensitivity analysis were conducted.

RESULTS

After a 7-year follow-up, 103 (1.61 %) of the 6400 participants developed asthma. Each 10 μg/m³ increment in the 1-, 2-, 3-, and 4-year moving average concentrations of PM_1-2.5 corresponded to a 1.82 [95 % confidence interval (CI):1.11-2.98], 1.95 (95 % CI: 1.24-3.07), 1.95 (95 % CI: 1.26-3.03) and 1.88 (95 % CI: 1.26-2.81) fold risk for incident asthma, respectively. A significant multiplicative interaction was observed between socioeconomic level and long-term exposure to PM_1-2.5. Stratified analysis showed that smokers and those with lower socioeconomic levels were at higher risk of incident asthma related to PM_1-2.5. Restricted cubic splines showed an increasing trend in asthma incidence with increasing PM_1-2.5. Sensitivity analyses showed that our model was robust.

CONCLUSION

Long-term exposure to PM_1-2.5 was positively associated with incident asthma in middle-aged and elderly individuals. Participants with a history of smoking and lower socioeconomic levels had a higher risk. More studies are warranted warrant to establish an accurate reference value of PM_1-2.5 to mitigate the growing asthma burden.

Association between Changes in Protein Intake and Risk of Cognitive Impairment: A Prospective Cohort Study

Little is known about the role of change in protein intake in affecting cognitive function among older adults. Therefore, we aimed to investigate the associations between the change in protein intake from various food groups and cognitive impairment among older adults in a prospective cohort study. A total of 6951 participants without cognitive impairment or dementia were included in this study. The frequency of protein intake from various food groups was measured by a food frequency questionnaire at baseline and follow-up. Multivariable Cox hazard models with time as the underlying time metric applied to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs). During the 37,535 person-years of follow-up, 1202 (17.3%) participants developed cognitive impairment. The improvement in overall protein intake was negatively associated with cognitive impairment with multivariable-adjusted HR of 0.98 (95% CI = 0.97-0.99). Compared with participants with stable change, those with an extreme decline in animal-based protein intake had a 48% higher risk of cognitive impairment. The associations of changes in protein from six food groups with cognitive impairment were in a similar direction to the main result. Protective associations between improving protein intake and a reduced risk of cognitive impairment were observed.

Association of leisure activity changes and reversion from mild cognitive impairment to normal cognitive function among older adults: A prospective cohort study

Background

Older adults with mild cognitive impairment (MCI) have the possibility of reverting to normal cognitive function. Leisure activity engagement (LAE) plays a critical role in the progress of the cognitive function. A better understanding of the dynamic relationship between LAE and MCI reversion would inform the implementation of preclinical dementia interventions. This study aimed to investigate the association between change patterns of LAE and MCI reversion among older adults using the Chinese Longitudinal Healthy Longevity Survey (CLHLS) database.

Study design

Longitudinal population-based study.

Methods

Older adults with MCI at the baseline were enrolled in this study. Information about cognitive function, overall, cognitively stimulating, physically active/demanding, and socially engaged LAE was collected at baseline and follow-up. Adjusted hazard ratios (HRs) for reversion and 95% confidence intervals (CIs) were calculated by Cox hazard models with time as the underlying time metric. We also assessed potential effect modifications by creating a cross-product of the stratifying variable with LAE change patterns in the fully adjusted model.

Results

The restricted cubic spline showed that the association between LAE change scores and MCI reversion rate was statistically significant and nonlinear (p<0.01). Taking participants in the low-low group as a reference, participants in the low-medium, low-high, medium-medium, medium-high, high-medium, and high-high groups had increased possibilities of MCI reversion with HRs (95% CI) of 2.19 (1.57-3.06), 2.97 (2.13-4.13), 0.87 (0.64-1.19), 2.28 (1.71-3.03), 2.78 (2.10-3.69), 1.93 (1.43-2.59), and 2.74 (2.09-3.60), respectively. Further stratified models showed that the impact of LAE change patterns on MCI reversion varied in different ages (nonagenarian, octogenarian, and younger elderly) and gender.

Conclusions

Participants who maintained the highest LAE had the greatest possibility of MCI reversion. Meanwhile, a higher level of LAE maintenance was associated with the increased possibility of MCI reversion. These results provide a practical message to older adults about how dynamic changes in LAE are associated with improved cognitive function.

Adherence to High Dietary Diversity and Incident Cognitive Impairment for the Oldest-Old: A Community-Based, Nationwide Cohort Study

BACKGROUND AND AIMS

Dietary diversity change is associated with cognitive function, however, whether the effect still exists among the oldest-old (80+) is unclear. Our aim was to examine the effect of dietary diversity changes on cognitive impairment for the oldest-old in a large prospective cohort.

METHODS

Within the Chinese Longitudinal Healthy Longevity Study, 6237 adults older than 80 years were included. The dietary diversity score (DDS) was assessed by a simplified food frequency questionnaire (FFQ). Cognitive impairment was defined as a Mini-Mental State Examination (MMSE) score lower than 18 points. Cognitive decline was defined as a reduction of total MMSE score ≥3 points, and cognitive decline of different subdomains was defined as a reduction of ≥15% in the corresponding cognitive domain. The multivariate-adjusted Cox proportional hazard model evaluated the effects of DDS change on cognitive decline. The linear mixed-effect model was used to test subsequent changes in MMSE over the years.

RESULTS

During 32,813 person-years of follow-up, 1829 participants developed cognitive impairment. Relative to the high-high DDS change pattern, participants in the low-low and high-low patterns were associated with an increased risk of cognitive impairment with a hazard ratio (95% confidential interval, CI) of 1.43 (1.25, 1.63) and 1.44 (1.24, 1.67), and a faster decline in the MMSE score over the follow-up year. Participants with the low-high pattern had a similar incidence of cognitive impairment with HRs (95% CI) of 1.03 (0.88, 1.20). Compared with the stable DDS status group (-1-1), the risk of cognitive impairment was higher for those with large declines in DDS (≤-5) and the HR was 1.70 (95% CI: 1.44, 2.01).

CONCLUSIONS

Even for people older than 80, dietary diversity change is a simple method to identify those who had a high risk of cognitive decline. Keeping high dietary diversity is beneficial for cognitive function and its subdomain even in the final phase of life, especially for females and the illiterate oldest-old.

Patient outcomes following AKI and AKD: a population-based cohort study

BACKGROUND

Acute kidney injury (AKI) is common and associated with adverse outcomes as well as important healthcare costs. However, evidence examining the epidemiology of acute kidney disease (AKD)-recently defined as AKI persisting between 7 and 90 days-remains limited. The aims of this study were to establish the rates of early AKI recovery, progression to AKD and non-recovery; examine risk factors associated with non-recovery and investigate the association between recovery timing and adverse outcomes, in a population-based cohort.

METHODS

All adult residents of Tayside & Fife, Scotland, UK, with at least one episode of community or hospital-managed AKI using KDIGO creatinine-based definition during the period 1 January 2010 to 31 December 2018 were identified. Logistic regression was used to examine factors associated with non-recovery, and Cox modelling was used to establish associations between AKI recovery timing and risks of mortality and development of de novo CKD.

RESULTS

Over 9 years, 56,906 patients with at least one AKI episode were identified with 18,773 (33%) of these progressing to AKD. Of those progressing to AKD, 5059 (27%) had still not recovered at day 90 post AKI diagnosis. Risk factors for AKD included: increasing AKI severity, pre-existing cancer or chronic heart failure and recent use of loop diuretics. Compared with early AKI recovery, progression to AKD was associated with increased hazard of 1-year mortality and de novo CKD (HR = 1.20, 95% CI 1.13 to 1.26 and HR = 2.21, 95% CI 1.91 to 2.57 respectively).

CONCLUSIONS

These findings highlight the importance of early AKI recognition and management to avoid progression to AKD and long-term adverse outcomes.

Interplay Between Duration of Androgen Deprivation Therapy and External Beam Radiotherapy With or Without a Brachytherapy Boost for Optimal Treatment of High-risk Prostate Cancer: A Patient-Level Data Analysis of 3 Cohorts

IMPORTANCE

Radiotherapy combined with androgen deprivation therapy (ADT) is a standard of care for high-risk prostate cancer. However, the interplay between radiotherapy dose and the required minimum duration of ADT is uncertain.

OBJECTIVE

To determine the specific ADT duration threshold that provides a distant metastasis-free survival (DMFS) benefit in patients with high-risk prostate cancer receiving external beam radiotherapy (EBRT) or EBRT with a brachytherapy boost (EBRT+BT).

DESIGN, SETTINGS, AND PARTICIPANTS

This was a cohort study of 3 cohorts assembled from a multicenter retrospective study (2000-2013); a post hoc analysis of the Randomized Androgen Deprivation and Radiotherapy 03/04 (RADAR; 2003-2007) randomized clinical trial (RCT); and a cross-trial comparison of the RADAR vs the Deprivación Androgénica y Radio Terapía (Androgen Deprivation and Radiation Therapy; DART) 01/05 RCT (2005-2010). In all, the study analyzed 1827 patients treated with EBRT and 1108 patients treated with EBRT+BT from the retrospective cohort; 181 treated with EBRT and 203 with EBRT+BT from RADAR; and 91 patients treated with EBRT from DART. The study was conducted from October 15, 2020, to July 1, 2021, and the data analyses, from January 5 to June 15, 2021.

EXPOSURES

High-dose EBRT or EBRT+BT for an ADT duration determined by patient-physician choice (retrospective) or by randomization (RCTs).

MAIN OUTCOMES AND MEASURES

The primary outcome was DMFS; secondary outcome was overall survival (OS). Natural cubic spline analysis identified minimum thresholds (months).

RESULTS

This cohort study of 3 studies totaling 3410 men (mean age [SD], 68 [62-74] years; race and ethnicity not collected) with high-risk prostate cancer found a significant interaction between the treatment type (EBRT vs EBRT+BT) and ADT duration (binned to <6, 6 to <18, and ≥18 months). Natural cubic spline analysis identified minimum duration thresholds of 26.3 months (95% CI, 25.4-36.0 months) for EBRT and 12 months (95% CI, 4.9-36.0 months) for EBRT+BT for optimal effect on DMFS. In RADAR, the prolongation of ADT for patients receiving only EBRT was not associated with significant improvements in DMFS (hazard ratio [HR], 1.01; 95% CI, 0.65-1.57); however, for patients receiving EBRT+BT, a longer duration was associated with improved DMFS (DMFS HR, 0.56; 95% CI, 0.36-0.87; P = .01). For patients receiving EBRT alone (DART), 28 months of ADT was associated with improved DMFS compared with 18 months (RADAR HR, 0.37; 95% CI, 0.17-0.80; P = .01).

CONCLUSIONS AND RELEVANCE

These cohort study findings suggest that the optimal minimum ADT duration for treatment with high-dose EBRT alone is more than 18 months; and for EBRT+BT, it is 18 months or possibly less. Additional studies are needed to determine more precise minimum durations.

K-means for shared frailty models

Background

The purpose of this research was to see how the k-means algorithm can be applied to survival analysis with single events per subject for defining groups, which can then be modeled in a shared frailty model to further allow the capturing the unmeasured confounding not already explained by the covariates in the model.

Methods

For this purpose we developed our own k-means survival grouping algorithm to handle this approach. We compared a regular shared frailty model with a regular grouping variable and a shared frailty model with a k-means grouping variable in simulations as well as analysis on a real dataset.

Results

We found that in both simulations as well as real data showed that our k-means clustering is no different than the typical frailty clustering even under different situations of varied case rates and censoring. It appeared our k-means algorithm could be a trustworthy mechanism of creating groups from data when no grouping term exists for including in a frailty term in a survival model or comparing to an existing grouping variable available in the current data to use in a frailty model.

The associations of blood pressure parameters with all-cause and cardiovascular mortality in peritoneal dialysis patients: a cohort study in China

BACKGROUND

It remains controversial to claim blood pressure (BP) as a leading risk factor for high risk of death in peritoneal dialysis patients, and less is known about the relationship between BP and mortality in Chinese peritoneal dialysis patients.

METHODS

From Zhejiang Renal Data System in China, we collected data on patients treated and followed up at 98 peritoneal dialysis centres from 2008 to 2016. The associations of BP parameters [SBP, DBP, mean arterial pressure (MAP) and pulse pressure (PP)] with all-cause and cardiovascular mortality were examined. We fitted Cox models for mortality with penalized splines using nonparametric smoothers. Several sensitivity analyses were performed to confirm the robustness of our primary findings.

RESULTS

A total of 7335 Chinese peritoneal dialysis patients were included. During a median follow-up of 35.8 months, 1281 (17.5%) patients died. SBP, DBP, MAP follow a U-shaped pattern of both all-cause and cardiovascular mortality. PP presents a reverse L-shaped association with all-cause mortality. Either a higher (SBP >141, DBP >85 or MAP >102 mmHg) or lower (SBP <119, DBP <67 or MAP <88 mmHg) BP tends to have a significantly higher all-cause and cardiovascular mortality risk. Higher PP (>60 mmHg) is related to a higher risk of all-cause mortality, but not cardiovascular mortality. These associations remain the same in our competing risk analysis and subgroup analyses.

CONCLUSION

These data indicate U-shaped associations of SBP, DBP and MAP with all-cause mortality and cardiovascular mortality, respectively, and a reverse L-shaped association of PP with all-cause mortality. Further studies are needed to reliably establish the optimal BP targets for better hypertension control in peritoneal dialysis patients.

Association of dietary diversity changes and mortality among older people: A prospective cohort study

BACKGROUND & AIMS

The association between dietary diversity (DD) changes and mortality remains unclear. We aimed to investigate the association between DD changes and all-cause mortality among older people.

METHODS

A total of 17,959 participants with a mean age of 84.8 years old were enrolled at baseline. Food groups were collected at baseline and follow-up using simplified food frequency questionnaire (FFQ), and then overall, plant-based and animal-based dietary diversity score (DDS) were calculated. DDS changes were calculated using DDS at baseline and the first follow-up. The association between three DDS changes (overall, plant-based and animal-based DDS) and subsequent all-cause mortality were evaluated. Nonparametrically restricted cubic splines and a multivariable-adjusted Cox proportional hazards model were used to estimate HRs and 95% CIs.

RESULTS

We documented 12,974 deaths over a 129,590 person-years of follow up. Compared with high-to-high DDS pattern, participants with lower overall DDS patterns had increased mortality risk with HRs (95%CI) of 1.39 (1.29-1.49), 1.53 (1.37-1.70), 1.38 (1.18-1.60) and 1.55 (1.31-1.83) for medium-to-medium, low-to-low, low-to-high and high-to-low patterns, respectively. And compared with high-to-high DDS pattern, the estimates were 1.34 (1.23-1.46), 1.49 (1.35-1.65), 1.43 (1.23-1.67) and 1.62 (1.40-1.88) for plant-based DDS, and 1.23 (1.15-1.31), 1.29 (1.20-1.40), 1.24 (1.12-1.37) and 1.28 (1.15-1.44) for animal-based DDS for medium-to-medium, low-to-low, low-to-high and high-to-low patterns, respectively. There was a U-shaped association between DDS change scores and mortality, and compared with participants with whose DDS remained stable, those with extreme declines and extreme improvements had higher risks of mortality with HRs (95% CI) of 1.15 (1.09-1.22) and 1.11 (1.04-1.17).

CONCLUSIONS

Maintaining a lower DDS, extreme declines and extreme improvements in DDS were all associated with an increased risk of all-cause mortality.

Disorders in bone-mineral parameters and the risk of death in persons with chronic kidney disease stages 4 and 5: the PECERA study

BACKGROUND

Abnormalities of bone mineral parameters are associated with increased mortality in patients on dialysis, but their effects and the optimal range of these biomarkers are less well characterized in non-dialysis chronic kidney disease (CKD).

METHODS

PECERA (Collaborative Study Project in Patients with Advanced CKD) is a 3-year, prospective multicenter, open-cohort study of 966 adult patients with non-dialyzed CKD stages 4-5 enrolled from 12 centers in Spain. Associations between levels of serum calcium (Ca) (corrected for albumin), phosphate (P), and intact parathyroid hormone (iPTH) with all-cause mortality (primary outcome) and cardiovascular mortality (secondary outcome) were examined using time-dependent Cox proportional hazards models and penalized splines analysis adjusted by demographics and comorbidities, treatments and biochemical values collected every 6 months for 3 years.

RESULTS

After a median follow-up of 29 months (IQR: 13-36 months) there were 181 deaths (19%). The association of calcium with all-cause mortality was J-shaped, with an increased risk for all-cause mortality at levels > 10.5 mg/dL. For phosphate and iPTH levels, the association was U-shaped. The serum values associated with the minimum risk of mortality were 3.8 mg/dL for phosphate and 70 pg/mL for iPTH, being the lowest risk ranges between 2.8 and 5.0 mg/dL, and between 38 and 112 pg/mL for phosphate and iPTH, respectively.

CONCLUSIONS

Our study provides evidence on the non-linear association of serum calcium, phosphate and iPTH levels with mortality in stage 4 and 5 CKD patients, and suggests potential survival benefits for controlling bone mineral parameters in this population, as previously reported for dialysis patients.

Higher dietary diversity scores and protein-rich food consumption were associated with lower risk of all-cause mortality in the oldest old

BACKGROUND & AIMS

Dietary diversity is widely advocated in national and international recommendations although whether the beneficial effects on survival or longevity still apply in the final phase of the lifespan remains understudied. We aimed to prospectively examine the association of dietary diversity, food items with all-cause mortality among the oldest old (80+) and determine whether dietary diversity recommendations were appropriate for this population.

METHODS

The study included 28,790 participants aged 80+ (9957 octogenarians, 9925 nonagenarians, and 8908 centenarians). A baseline dietary diversity score (DDS) was constructed based on nine food items of a food frequency questionnaire. Cox models with penalized splines evaluated non-linear associations of DDS as continuous variable with mortality to identify cut-offs of DDS.

RESULTS

We documented 23,503 deaths during 96,739 person-years of follow-up. Each one unit increase in DDS was associated with a 9% lower risk of mortality (adjusted hazard ratio (HR): 0.91; 95% confidential interval (CI): 0.90-0.92). Compared to participants whose DDS less than 2 scores, those with a DDS of 2, 3, 4, 5, and higher than 6 scores had a lower mortality risk, the HRs were 0.86 (0.82-0.89), 0.78 (0.75-0.81), 0.69 (0.66-0.72), 0.65 (0.62-0.68), and 0.56 (0.53-0.58) respectively, and a significant trend emerged (p < 0.001). Protein-rich food items were associated with prominent beneficial effects on mortality including meat (HR and 95% CI for high vs low frequency: 0.70 (0.68-0.72)), fish and sea food (HR, 0.74 (0.72-0.77)), egg (HR, 0.75 (0.73-0.77)), and bean (HR, 0.80 (0.78-0.82)).

CONCLUSIONS

Even after the age of 80, the DDS tool may offer a simple and straightforward mean of identifying and screening individuals at high risk for mortality. Recommendation of dietary diversity, especially consumption of protein-rich food, may be advocated to reduce mortality risk and promote longevity in the oldest old.

Glycated Hemoglobin and All-Cause and Cause-Specific Mortality Among Adults With and Without Diabetes

CONTEXT

The patterns of associations between glycated Hb (HbA1c) and mortality are still unclear.

OBJECTIVE

To explore the extent to which ranges of HbA1c levels are associated with the risk of mortality among participants with and without diabetes.

DESIGN, SETTING, AND PATIENTS

This was a nationwide, community-based prospective cohort study. Included were 15,869 participants (median age 64 years) of the Health and Retirement Study, with available HbA1c data and without a history of cancer. Cox proportional hazards regression models were used to estimate hazard ratios with 95% CIs for mortality.

RESULTS

A total of 2133 participants died during a median follow-up of 5.8 years. In participants with diabetes, those with an HbA1c level of 6.5% were at the lowest risk of all-cause mortality. When HbA1c level was <5.6% or >7.4%, the increased all-cause mortality risk became statistically significant as compared with an HbA1c level of 6.5%. As for participants without diabetes, those with an HbA1c level of 5.4% were at the lowest risk of all-cause mortality. When the HbA1c level was <5.0%, the increased all-cause mortality risk became statistically significant as compared with an HbA1c level of 5.4%. However, we did not observe a statistically significant elevated risk of all-cause mortality above an HbA1c level of 5.4%.

CONCLUSIONS

A U-shaped and reverse J-shaped association for all-cause mortality was found among participants with and without diabetes. The corresponding optimal ranges for overall survival are predicted to be 5.6% and 7.4% and 5.0% and 6.5%, respectively.

Revisiting the association of blood pressure with mortality in oldest old people in China: community based, longitudinal prospective study

OBJECTIVE

To examine the associations of blood pressure with all cause mortality and cause specific mortality at three years among oldest old people in China.

DESIGN

Community based, longitudinal prospective study.

SETTING

2011 and 2014 waves of the Chinese Longitudinal Healthy Longevity Survey, conducted in 22 Chinese provinces.

PARTICIPANTS

4658 oldest old individuals (mean age 92.1 years).

MAIN OUTCOME MEASURES

All cause mortality and cause specific mortality assessed at three year follow-up.

RESULTS

1997 deaths were recorded at three year follow-up. U shaped associations of mortality with systolic blood pressure, mean arterial pressure, and pulse pressure were identified; values of 143.5 mm Hg, 101 mm Hg, and 66 mm Hg conferred the minimum mortality risk, respectively. After adjustment for covariates, the U shaped association remained only for systolic blood pressure (minimum mortality risk at 129 mm Hg). Compared with a systolic blood pressure value of 129 mm Hg, risk of all cause mortality decreased for values lower than 107 mm Hg (from 1.47 (95% confidence interval 1.01 to 2.17) to 1.08 (1.01 to 1.17)), and increased for values greater than 154 mm Hg (from 1.08 (1.01 to 1.17) to 1.27 (1.02 to 1.58)). In the cause specific analysis, compared with a middle range of systolic blood pressure (107-154 mm Hg), higher values (>154 mm Hg) were associated with a higher risk of cardiovascular mortality (adjusted hazard ratio 1.51 (95% confidence interval 1.12 to 2.02)); lower values (<107 mm Hg) were associated with a higher risk of non-cardiovascular mortality (1.58 (1.26 to 1.98)). The U shaped associations remained in sensitivity and subgroup analyses.

CONCLUSIONS

This study indicates a U shaped association between systolic blood pressure and all cause mortality at three years among oldest old people in China. This association could be explained by the finding that higher systolic blood pressure predicted a higher risk of death from cardiovascular disease, and that lower systolic blood pressure predicted a higher risk of death from non-cardiovascular causes. These results emphasise the importance of revisiting blood pressure management or establishing specific guidelines for management among oldest old individuals.

Complete hazard ranking to analyze right-censored data: An ALS survival study

Survival analysis represents an important outcome measure in clinical research and clinical trials; further, survival ranking may offer additional advantages in clinical trials. In this study, we developed GuanRank, a non-parametric ranking-based technique to transform patients' survival data into a linear space of hazard ranks. The transformation enables the utilization of machine learning base-learners including Gaussian process regression, Lasso, and random forest on survival data. The method was submitted to the DREAM Amyotrophic Lateral Sclerosis (ALS) Stratification Challenge. Ranked first place, the model gave more accurate ranking predictions on the PRO-ACT ALS dataset in comparison to Cox proportional hazard model. By utilizing right-censored data in its training process, the method demonstrated its state-of-the-art predictive power in ALS survival ranking. Its feature selection identified multiple important factors, some of which conflicts with previous studies.

Estimating and Interpreting Effects from Nonlinear Exposure-Response Curves in Occupational Cohorts Using Truncated Power Basis Expansions and Penalized Splines

Truncated power basis expansions and penalized spline methods are demonstrated for estimating nonlinear exposure-response relationships in the Cox proportional hazards model. R code is provided for fitting models to get point and interval estimates. The method is illustrated using a simulated data set under a known exposure-response relationship and in a data application examining risk of carpal tunnel syndrome in an occupational cohort.

Classification of Airflow Limitation Based on z-Score Underestimates Mortality in Patients with Chronic Obstructive Pulmonary Disease

RATIONALE

Global Lung Function Initiative recommends reporting lung function measures as z-score, and a classification of airflow limitation (AL) based on this parameter has recently been proposed.

OBJECTIVES

To evaluate the prognostic capacity of the AL classifications based on z-score or percentage predicted of FEV₁ in patients with chronic obstructive pulmonary disease (COPD).

METHODS

A cohort of 2,614 patients with COPD recruited outside the hospital setting was examined after a mean (± SD) of 57 ± 13 months of follow-up, totaling 10,322 person-years. All-cause mortality was analyzed, evaluating the predictive capacity of several AL staging systems.

MEASUREMENTS AND MAIN RESULTS

Based on Global Initiative for Chronic Obstructive Lung Disease guidelines, 461 patients (17.6%) had mild, 1,452 (55.5%) moderate, 590 (22.6%) severe, and 111 (4.2%) very severe AL. According to z-score classification, 66.3% of patients remained with the same severity, whereas 23.7% worsened and 10.0% improved. Unlike other staging systems, patients with severe AL according to z-score had higher mortality than those with very severe AL (increase of risk by 5.2 and 3.9 times compared with mild AL, respectively). The predictive capacity for 5-year survival was slightly higher for FEV₁ expressed as percentage of predicted than as z-score (area under the curve: 0.714-0.760 vs. 0.649-0.708, respectively). A severity-dependent relationship between AL grades by z-score and mortality was only detected in patients younger than age 60 years.

CONCLUSIONS

In patients with COPD, the AL classification based on z-score predicts worse mortality than those based on percentage of predicted. It is possible that the z-score underestimates AL severity in patients older than 60 years of age with severe functional impairment.

Metabolic Acidosis and Long-Term Clinical Outcomes in Kidney Transplant Recipients

Metabolic acidosis (MA), indicated by low serum total CO₂ (TCO₂) concentration, is a risk factor for mortality and progressive renal dysfunction in CKD. However, the long-term effects of MA on kidney transplant recipients (KTRs) are unclear. We conducted a multicenter retrospective cohort study of 2318 adult KTRs, from January 1, 1997 to March 31, 2015, to evaluate the prevalence of MA and the relationships between TCO₂ concentration and clinical outcomes. The prevalence of low TCO₂ concentration (<22 mmol/L) began to increase in KTRs with eGFR<60 ml/min per 1.73 m² and ranged from approximately 30% to 70% in KTRs with eGFR<30 ml/min per 1.73 m² Multivariable Cox proportional hazards models revealed that low TCO₂ concentration 3 months after transplant associated with increased risk of graft loss (hazard ratio [HR], 1.74%; 95% confidence interval [95% CI], 1.26 to 2.42) and death-censored graft failure (DCGF) (HR, 1.66; 95% CI, 1.14 to 2.42). Cox regression models using time-varying TCO₂ concentration additionally demonstrated significant associations between low TCO₂ concentration and graft loss (HR, 3.48; 95% CI, 2.47 to 4.90), mortality (HR, 3.16; 95% CI, 1.77 to 5.62), and DCGF (HR, 3.17; 95% CI, 2.12 to 4.73). Marginal structural Cox models adjusted for time-varying eGFR further verified significant hazards of low TCO₂ concentration for graft loss, mortality, and DCGF. In conclusion, MA was frequent in KTRs despite relatively preserved renal function and may be a significant risk factor for graft failure and patient mortality, even after adjusting for eGFR.

Comparing Smoothing Techniques for Fitting the Nonlinear Effect of Covariate in Cox Models

BACKGROUND AND OBJECTIVE

Cox model is a popular model in survival analysis, which assumes linearity of the covariate on the log hazard function, While continuous covariates can affect the hazard through more complicated nonlinear functional forms and therefore, Cox models with continuous covariates are prone to misspecification due to not fitting the correct functional form for continuous covariates. In this study, a smooth nonlinear covariate effect would be approximated by different spline functions.

MATERIAL AND METHODS

We applied three flexible nonparametric smoothing techniques for nonlinear covariate effect in the Cox models: penalized splines, restricted cubic splines and natural splines. Akaike information criterion (AIC) and degrees of freedom were used to smoothing parameter selection in penalized splines model. The ability of nonparametric methods was evaluated to recover the true functional form of linear, quadratic and nonlinear functions, using different simulated sample sizes. Data analysis was carried out using R 2.11.0 software and significant levels were considered 0.05.

RESULTS

Based on AIC, the penalized spline method had consistently lower mean square error compared to others to selection of smoothed parameter. The same result was obtained with real data.

CONCLUSION

Penalized spline smoothing method, with AIC to smoothing parameter selection, was more accurate in evaluate of relation between covariate and log hazard function than other methods.

Flexible regression models for rate differences, risk differences and relative risks

Generalized additive models (GAMs) based on the binomial and Poisson distributions can be used to provide flexible semi-parametric modelling of binary and count outcomes. When used with the canonical link function, these GAMs provide semi-parametrically adjusted odds ratios and rate ratios. For adjustment of other effect measures, including rate differences, risk differences and relative risks, non-canonical link functions must be used together with a constrained parameter space. However, the algorithms used to fit these models typically rely on a form of the iteratively reweighted least squares algorithm, which can be numerically unstable when a constrained non-canonical model is used. We describe an application of a combinatorial EM algorithm to fit identity link Poisson, identity link binomial and log link binomial GAMs in order to estimate semi-parametrically adjusted rate differences, risk differences and relative risks. Using smooth regression functions based on B-splines, the method provides stable convergence to the maximum likelihood estimates, and it ensures that the estimates always remain within the parameter space. It is also straightforward to apply a monotonicity constraint to the smooth regression functions. We illustrate the method using data from a clinical trial in heart attack patients.

Mortality prediction in chronic obstructive pulmonary disease comparing the GOLD 2007 and 2011 staging systems: a pooled analysis of individual patient data

BACKGROUND

There is no universal consensus on the best staging system for chronic obstructive pulmonary disease (COPD). Although documents (eg, the Global Initiative for Chronic Obstructive Lung Disease [GOLD] 2007) have traditionally used forced expiratory volume in 1 s (FEV1) for staging, clinical parameters have been added to some guidelines (eg, GOLD 2011) to improve patient management. As part of the COPD Cohorts Collaborative International Assessment (3CIA) initiative, we aimed to investigate how individual patients were categorised by GOLD 2007 and 2011, and compare the prognostic accuracy of the staging documents for mortality.

METHODS

We searched reports published from Jan 1, 2008, to Dec 31, 2014. Using data from cohorts that agreed to participate and had a minimum amount of information needed for GOLD 2007 and 2011, we did a patient-based pooled analysis of existing data. With use of raw data, we recalculated all participant assignments to GOLD 2007 I-IV classes, and GOLD 2011 A-D stages. We used survival analysis, C statistics, and non-parametric regression to model time-to-death data and compare GOLD 2007 and GOLD 2011 staging systems to predict mortality.

FINDINGS

We collected individual data for 15 632 patients from 22 COPD cohorts from seven countries, totalling 70 184 person-years. Mean age of the patients was 63·9 years (SD 10·1); 10 751 (69%) were men. Based on FEV1 alone (GOLD 2007), 2424 (16%) patients had mild (I), 7142 (46%) moderate (II), 4346 (28%) severe (III), and 1670 (11%) very severe (IV) disease. We compared staging with the GOLD 2007 document with that of the new GOLD 2011 system in 14 660 patients: 5548 (38%) were grade A, 2733 (19%) were grade B, 1835 (13%) were grade C, and 4544 (31%) were grade D. GOLD 2011 shifted the overall COPD severity distribution to more severe categories. There were nearly three times more COPD patients in stage D than in former stage IV (p<0·05). The predictive capacity for survival up to 10 years was significant for both systems (p<0·01) but area under the curves were only 0·623 (GOLD 2007) and 0·634 (GOLD 2011), and GOLD 2007 and 2011 did not differ significantly. We identified the percent predicted FEV1 thresholds of 85%, 55% and 35% as better to stage COPD severity for mortality, which are similar to the ones used previously.

INTERPRETATION

Neither GOLD COPD classification schemes have sufficient discriminatory power to be used clinically for risk classification at the individual level to predict total mortality for 3 years of follow-up and onwards. Increasing intensity of treatment of patients with COPD due to their GOLD 2011 reclassification is not known to improve health outcomes. Evidence-based thresholds should be searched when exploring the prognostic ability of current and new COPD multicomponent indices.

FUNDING

None.

Improvement of mineral and bone metabolism markers is associated with better survival in haemodialysis patients: the COSMOS study

BACKGROUND

Abnormalities in serum phosphorus, calcium and parathyroid hormone (PTH) have been associated with poor survival in haemodialysis patients. This COSMOS (Current management Of Secondary hyperparathyroidism: a Multicentre Observational Study) analysis assesses the association of high and low serum phosphorus, calcium and PTH with a relative risk of mortality. Furthermore, the impact of changes in these parameters on the relative risk of mortality throughout the 3-year follow-up has been investigated.

METHODS

COSMOS is a 3-year, multicentre, open-cohort, prospective study carried out in 6797 adult chronic haemodialysis patients randomly selected from 20 European countries.

RESULTS

Using Cox proportional hazard regression models and penalized splines analysis, it was found that both high and low serum phosphorus, calcium and PTH were associated with a higher risk of mortality. The serum values associated with the minimum relative risk of mortality were 4.4 mg/dL for serum phosphorus, 8.8 mg/dL for serum calcium and 398 pg/mL for serum PTH. The lowest mortality risk ranges obtained using as base the previous values were 3.6-5.2 mg/dL for serum phosphorus, 7.9-9.5 mg/dL for serum calcium and 168-674 pg/mL for serum PTH. Decreases in serum phosphorus and calcium and increases in serum PTH in patients with baseline values of >5.2 mg/dL (phosphorus), >9.5 mg/dL (calcium) and <168 pg/mL (PTH), respectively, were associated with improved survival.

CONCLUSIONS

COSMOS provides evidence of the association of serum phosphorus, calcium and PTH and mortality, and suggests survival benefits of controlling chronic kidney disease-mineral and bone disorder biochemical parameters in CKD5D patients.

Data generation for the Cox proportional hazards model with time-dependent covariates: a method for medical researchers

The proliferation of longitudinal studies has increased the importance of statistical methods for time-to-event data that can incorporate time-dependent covariates. The Cox proportional hazards model is one such method that is widely used. As more extensions of the Cox model with time-dependent covariates are developed, simulations studies will grow in importance as well. An essential starting point for simulation studies of time-to-event models is the ability to produce simulated survival times from a known data generating process. This paper develops a method for the generation of survival times that follow a Cox proportional hazards model with time-dependent covariates. The method presented relies on a simple transformation of random variables generated according to a truncated piecewise exponential distribution and allows practitioners great flexibility and control over both the number of time-dependent covariates and the number of time periods in the duration of follow-up measurement. Within this framework, an additional argument is suggested that allows researchers to generate time-to-event data in which covariates change at integer-valued steps of the time scale. The purpose of this approach is to produce data for simulation experiments that mimic the types of data structures applied that researchers encounter when using longitudinal biomedical data. Validity is assessed in a set of simulation experiments, and results indicate that the proposed procedure performs well in producing data that conform to the assumptions of the Cox proportional hazards model.

Fruit and vegetable intake and the risk of colorectal cancer: results from the Shanghai Men's Health Study

PURPOSE

The observed associations of fruit and vegetable consumption with the risk of colorectal cancer have been inconsistent. Therefore, we aimed to evaluate the association of fruit and vegetable consumption with the risk of colorectal cancer among Chinese men.

METHODS

61,274 male participants aged 40-74 years were included. A validated food frequency questionnaire was administered to collect information on usual dietary intake, including 8 fruits and 38 vegetables commonly consumed by residents of Shanghai. Follow-up for diagnoses of colon or rectal cancer was available through 31 December 2010. Dietary intakes were analyzed both as categorical and continuous variables. Multivariable-adjusted hazard ratios (HRs) and 95 % confidence intervals (95 % CIs) were calculated for colorectal, colon, and rectal cancers using Cox proportional hazards models.

RESULTS

After 390,688 person-years of follow-up, 398 cases of colorectal cancer (236 colon and 162 rectal) were observed in the cohort. Fruit consumption was inversely associated with the risk of colorectal cancer (fifth vs. first quintile HR 0.67; 95 % CI 0.48, 0.95; p trend = 0.03), whereas vegetable intake was not significantly associated with risk. The associations for subgroups of fruits and legumes, but not other vegetable categories, were generally inversely associated with the risk of colon and rectal cancers.

CONCLUSIONS

Fruit intake was generally inversely associated with the risk of colorectal cancer, whereas vegetable consumption was largely unrelated to risk among middle-aged and older Chinese men.

Seasonality of viral infections: mechanisms and unknowns

Seasonality is a long-recognized attribute of many viral infections of humans, but the mechanisms underlying seasonality, particularly for person-to-person communicable diseases, remain poorly understood. Better understanding of drivers of seasonality could provide insights into the relationship between the physical environment and infection risk, which is particularly important in the context of global ecological change in general, and climate change in particular. In broad terms, seasonality represents oscillation in pathogens' effective reproductive number, which, in turn, must reflect oscillatory changes in infectiousness, contact patterns, pathogen survival, or host susceptibility. Epidemiological challenges to correct identification of seasonal drivers of risk include failure to adjust for predictable correlation between disease incidence and seasonal exposures, and unmeasured confounding. The existing evidence suggests that the seasonality of some enteric and respiratory viral pathogens may be driven by enhanced wintertime survival of pathogens, and also by increased host susceptibility resulting from relative 'wintertime immune suppression'. For vector-borne diseases and zoonoses, environmental influences on vector or reservoir abundance, and vector biting rates, are probably more important. However, numerous areas of uncertainty exist, making this an exciting area for future research.