IDNetwork: A deep illness-death network based on multi-state event history process for disease prognostication.

Multi-state models can capture the different patterns of disease evolution. In particular, the illness-death model is used to follow disease progression from a healthy state to an intermediate state of the disease and to a death-related final state. We aim to use those models in order to adapt treatment decisions according to the evolution of the disease. In state-of-the art methods, the risks of transition between the states are modeled via (semi-) Markov processes and transition-specific Cox proportional hazard (P.H.) models. The Cox P.H. model assumes that each variable makes a linear contribution to the model, but the relationship between covariates and risks can be more complex in clinical situations. To address this challenge, we propose a neural network architecture called illness-death network (IDNetwork) that relaxes the linear Cox P.H. assumption within an illness-death process. IDNetwork employs a multi-task architecture and uses a set of fully connected subnetworks in order to learn the probabilities of transition. Through simulations, we explore different configurations of the architecture and demonstrate the added value of our model. IDNetwork significantly improves the predictive performance compared to state-of-the-art methods on a simulated data set, on two clinical trials for patients with colon cancer and on a real-world data set in breast cancer.

Prognostic value of high-sensitivity measurable residual disease assessment after front-line chemoimmunotherapy in chronic lymphocytic leukemia.

Measurable residual disease (MRD) status is widely adopted in clinical trials in patients with chronic lymphocytic leukemia (CLL). Findings from FILO group trials (CLL2007FMP, CLL2007SA, CLL2010FMP) enabled investigation of the prognostic value of high-sensitivity (0.7 × 10-5) MRD assessment using flow cytometry, in blood (N = 401) and bone marrow (N = 339), after fludarabine, cyclophosphamide, and rituximab (FCR)-based chemoimmunotherapy in a homogeneous population with long follow-up (median 49.5 months). Addition of low-level positive MRD < 0.01% to MRD ≥ 0.01% increased the proportion of cases with positive MRD in blood by 39% and in bone marrow by 27%. Compared to low-level positive MRD < 0.01%, undetectable MRD was associated with significantly longer progression-free survival (PFS) when using blood (72.2 versus 42.7 months; hazard ratio 0.40, p = 0.0003), but not when using bone marrow. Upon further stratification, positive blood MRD at any level, compared to undetectable blood MRD, was associated with shorter PFS irrespective of clinical complete or partial remission, and a lower 5-year PFS rate irrespective of IGHV-mutated or -unmutated status (all p < 0.05). In conclusion, high-sensitivity (0.0007%) MRD assessment in blood yielded additional prognostic information beyond the current standard sensitivity (0.01%). Our approach provides a model for future determination of the optimal MRD investigative strategy for any regimen.

Automated interpretation of home blood pressure assessment (Hy-Result software) versus physician's assessment: a validation study.

OBJECTIVE: Hy-Result is the first software for self-interpretation of home blood pressure measurement results, taking into account both the recommended thresholds for normal values and patient characteristics. We compare the software-generated classification with the physician's evaluation. DESIGN METHOD: The primary assessment criterion was whether algorithm classification of the blood pressure (BP) status concurred with the physician's advice (blinded to the software's results) following a consultation (n=195 patients). Secondary assessment was the reliability of text messages. RESULTS: In the 58 untreated patients, the agreement between classification of the BP status generated by the software and the physician's classification was 87.9%. In the 137 treated patients, the agreement was 91.9%. The κ-test applied for all the patients was 0.81 (95% confidence interval: 0.73-0.89). After correction of errors identified in the algorithm during the study, agreement increased to 95.4% [κ=0.9 (95% confidence interval: 0.84-0.97)]. For 100% of the patients with comorbidities (n=46), specific text messages were generated, indicating that a physician might recommend a target BP lower than 135/85 mmHg. Specific text messages were also generated for 100% of the patients for whom global cardiovascular risks markedly exceeded norms. CONCLUSION: Classification by Hy-Result is at least as accurate as that of a specialist in current practice (http://www.hy-result.com).

Banking or Bankrupting: Strategies for Sustaining the Economic Future of Public Cord Blood Banks.

BACKGROUND: Cord blood is an important source of stem cells. However, nearly 90% of public cord blood banks have declared that they are struggling to maintain their financial sustainability and avoid bankruptcy. The objective of this study is to evaluate how characteristics of cord blood units influence their utilization, then use this information to model the economic viability and therapeutic value of different banking strategies. METHODS: Retrospective analysis of cord blood data registered between January 1st, 2009 and December 31st, 2011 in Bone Marrow Donor Worldwide. Data were collected from four public banks in France, Germany and the USA. Samples were eligible for inclusion in the analysis if data on cord blood and maternal HLA typing and biological characteristics after processing were available (total nucleated and CD34+ cell counts). 9,396 banked cord blood units were analyzed, of which 5,815 were Caucasian in origin. A multivariate logistic regression model assessed the influence of three parameters on the CBU utilization rate: ethnic background, total nucleated and CD34+ cell counts. From this model, we elaborated a Utilization Score reflecting the probability of transplantation for each cord blood unit. We stratified three Utilization Score thresholds representing four different banking strategies, from the least selective (scenario A) to the most selective (scenario D). We measured the cost-effectiveness ratio for each strategy by comparing performance in terms of number of transplanted cord blood units and level of financial deficit. RESULTS: When comparing inputs and outputs over three years, Scenario A represented the most extreme case as it delivered the highest therapeutic value for patients (284 CBUs transplanted) along with the highest financial deficit (USD 5.89 million). We found that scenario C resulted in 219 CBUs transplanted with a limited deficit (USD 0.98 million) that charities and public health could realistically finance over the long term. We also found that using a pre-freezing level of 18 x 10(8) TNC would be the most cost-effective strategy for a public bank. CONCLUSION: Our study shows that a swift transition from strategy A to C can play a vital role in preventing public cord blood banks worldwide from collapsing.

Treatment for stable coronary artery disease: a network meta-analysis of cost-effectiveness studies.

INTRODUCTION AND OBJECTIVES: Numerous studies have assessed cost-effectiveness of different treatment modalities for stable angina. Direct comparisons, however, are uncommon. We therefore set out to compare the efficacy and mean cost per patient after 1 and 3 years of follow-up, of the following treatments as assessed in randomized controlled trials (RCT): medical therapy (MT), percutaneous coronary intervention (PCI) without stent (PTCA), with bare-metal stent (BMS), with drug-eluting stent (DES), and elective coronary artery bypass graft (CABG). METHODS: RCT comparing at least two of the five treatments and reporting clinical and cost data were identified by a systematic search. Clinical end-points were mortality and myocardial infarction (MI). The costs described in the different trials were standardized and expressed in US $ 2008, based on purchasing power parity. A network meta-analysis was used to compare costs. RESULTS: Fifteen RCT were selected. Mortality and MI rates were similar in the five treatment groups both for 1-year and 3-year follow-up. Weighted cost per patient however differed markedly for the five treatment modalities, at both one year and three years (P<0.0001). MT was the least expensive treatment modality: US $3069 and 13 864 after one and three years of follow-up, while CABG was the most costly: US $27 003 and 28 670 after one and three years. PCI, whether with plain balloon, BMS or DES came in between, but was closer to the costs of CABG. CONCLUSIONS: Appreciable savings in health expenditures can be achieved by using MT in the management of patients with stable angina.

Estimating and testing for center effects in competing risks.

The problems of fitting Gaussian frailties proportional hazards models for the subdistribution of a competing risk and of testing for center effects are considered. In the analysis of competing risks data, Fine and Gray proposed a proportional hazards model for the subdistribution to directly assess the effects of covariates on the marginal failure probabilities of a given failure cause. Katsahianbiet al. extended their model to clustered time to event data, by including random center effects or frailties in the subdistribution hazard. We first introduce an alternate estimation procedure to the one proposed by Katsahian et al. This alternate estimation method is based on the penalized partial likelihood approach often used in fitting Gaussian frailty proportional hazards models in the standard survival analysis context, and has the advantage of using standard survival analysis software. Second, four hypothesis tests for the presence of center effects are given and compared via Monte-Carlo simulations. Statistical and numerical considerations lead us to formulate pragmatic guidelines as to which of the four tests is preferable. We also illustrate the proposed methodology with registry data from bone marrow transplantation for acute myeloid leukemia (AML).

Risk assessment for buccal gingival recession defects in an adult population.

BACKGROUND: Little is known about systemic, environmental, and periodontal risk factors of gingival recessions. This study identifies variables related to buccal gingival recessions in an adult French population; this study uses data collected in the First National Periodontal and Systemic Examination Survey. METHODS: This cross-sectional survey includes 2,074 subjects (age range: 35 to 65 years) from a nationally representative sample obtained by a quota method stratified by age, gender, socioeconomic status, and geographic areas. The subjects had undergone a full-mouth periodontal examination, assessment of missing teeth, laboratory tests, and questionnaires. In the present analysis, the buccal gingival recession status of each subject was the outcome of interest and was assessed by criteria based on the severity and extent of gingival recessions. RESULTS: A total of 84.6% of the sample had at least one gingival recession. A multivariate linear regression model with backward selection showed that age (P >0.001), gender (P = 0.003), plaque index (P <0.001), and tobacco consumption (P <0.001) were associated with the extent of gingival recession. In addition to these variables, the number of missing teeth (P <0.001) and the gingival bleeding index (P = 0.010) were also associated with the severity of gingival recession. CONCLUSIONS: This study indicates that the risk factors for gingival recessions are similar to the traditional risk factors for periodontitis. However, the present model indicates that diabetes, increase of the body mass index, and alcohol intake are not associated with gingival recessions.

A comparison of methods of aesthetic assessment in root coverage procedures.

AIM: To evaluate the reliability of professional qualitative scoring methods used in evaluating aesthetic results after root coverage therapy and to evaluate the relationship between subjective and objective measurements. MATERIAL AND METHODS: A review panel of seven professional and non-professional, trained and untrained observers used photographic records to assess the overall cosmetic results of 162 root coverage surgical procedures in 133 patients (mean follow-up 17.51+/-17.37 months). Two different methods were used. In the before-after panel scoring system, observers evaluated the difference between preoperative and postoperative views, whereas in the random panel scoring system, observers rated each photograph independently. RESULTS: For both methods, intrarater agreement ranged from substantial to almost perfect for the periodontists. The best interrater agreement was found for trained periodontists using the five-point ordinal scale of the before-after panel scoring system (kappa=0.68). Neither root coverage percentage nor gingival augmentation was correlated to cosmetic assessment. CONCLUSIONS: The before-after scoring system is an acceptable and reliable method for professional cosmetic assessment of root coverage therapy. The overall cosmetic evaluation does not appear to be related to the percentage of root coverage.

Analysing multicentre competing risks data with a mixed proportional hazards model for the subdistribution.

In the competing-risks setting, to test the effect of a covariate on the probability of one particular cause of failure, the Fine and Gray model for the subdistribution hazard can be used. However, sometimes, competing risks data cannot be considered as independent because of a clustered design, for instance in registry cohorts or multicentre clinical trials. Frailty models have been shown useful to analyse such clustered data in a classical survival setting, where only one risk acts on the population. Inclusion of random effects in the subdistribution hazard has not been assessed yet. In this work, we propose a frailty model for the subdistribution hazard. This allows first to assess the heterogeneity across clusters, then to incorporate such an effect when testing the effect of a covariate of interest. Based on simulation study, the effect of the presence of heterogeneity on testing for covariate effects was studied. Finally, the model was illustrated on a data set from a registry cohort of patients with acute myeloid leukaemia who underwent bone marrow transplantation.

The graft-versus-leukaemia effect after allogeneic bone-marrow transplantation: assessment through competing risks approaches.

In failure time studies involving a chronic disease such as cancer, data often focus on one or more non-fatal events, in addition to survival, to describe the course of the disease. In the example of allogeneic bone marrow transplantation in leukaemic patients, acute graft-versus host disease (aGvHD), relapse and death were taken as the reported events, and we focused on testing the existence of graft-versus-leukaemia (GvL) effect, i.e. that the occurrence of aGvHD modifies the probability of relapse. One of the weaknesses of the standard competing risks models is their inability to model secondary relapses. We thus derived, from two competing risks models, two estimators of cumulative incidence functions of primary and secondary relapses, as well as statistics to test the GvL effect. The approach is illustrated by application to a data set from a cohort of 442 children with acute leukaemia who received an unrelated transplant.