Therneau, Terry M.

AbstractBackgroundData shared for research use must be de‐identified to protect participant confidentiality. A recent study has shown that face imagery from brain imaging studies could be used to re‐identify anonymous participants via automatic face recognition, linking them with all study‐collected protected health information. Here we measure how well existing software for removing MRI face imagery protects against this privacy threat.MethodWe recruited 153 participants with existing 3D FLAIR MRI and photographed their face from 5 varying angles. We used surf_ice to reconstruct each participant’s face from their MRI and tested whether free, publicly‐available face recognition software (Microsoft Azure) could match their photos to their correct MRI. We tested two existing programs for removing face voxels in MRI: mri_deface (FreeSurfer) and pydeface (python package), and one in‐house implementation. We observed that existing methods often left participants’ eyes intact while removing most of their nose/mouth/chin. This prevented Azure from detecting the reconstructions as faces, therefore preventing its attempt of face recognition. We hypothesized that the retained eye imagery alone could be sufficient for face recognition, if a face were detected. Therefore, we also reconstructed each de‐faced MRI after applying novel pre‐processing that replaced any removed/missing face voxels with those from our population‐average template, e.g. replacing missing noses/mouths/chins with population‐average noses/mouths/chins, before testing face recognition.ResultsWithout any MRI de‐facing, Microsoft Azure correctly matched 78% of participants’ photos to their MRI. After mri_deface and pydeface, 8% and 6% were still identified, respectively. After preprocessing the de‐faced MRI to replace missing areas with those from a population‐average template prior to face reconstruction, the face imagery remaining in the de‐faced images was enough to allow re‐identification of 23% and 20% of participants, respectively. With an in‐house face removal implementation that removes the whole face from every image, 0% of faces were re‐identified from the standard reconstructions, and 5% were identified from the “missing‐parts‐replacement” reconstructions.ConclusionsEven after applying popular MRI de‐facing software, Microsoft Azure face recognition automatically re‐identified up to 23% of participants. Better de‐facing software is needed to remove more of the face more reliably, without affecting brain measurements from de‐faced images.

AbstractBackgroundThe Network Failure Quotient (NFQ) metric summarizes functional MRI network integrity. We wanted to investigate the stability of the NFQ results across time, hardware, software and protocol changes during the last decade in ADNI Resting State (RS‐) fMRI.MethodIn ADNI‐2 (2009‐2016) RS‐fMRI was acquired on Philips scanners. In ADNI‐3 (2017‐present) RS‐fMRI is acquired on GE, Philips, and Siemens systems with “Basic” and “Advanced” variants based on the availability of simultaneous multi‐slice (SMS, multi‐band) imaging. (Table 1.) Resting state images were preprocessed by performing slice timing correction, trimming volumes from the first 10 seconds of acquired data, despiking and within‐series alignment. Simultaneous bandpass and regression was performed using spatial registration parameters and their first temporal derivatives, mean WM signal, mean CSF signal and global signal. Dual regression via GIFT was applied creating regional time courses. The NFQ is calculated as NFQ=(pDMN_to_vDMN + pDMN_to_adDMN) / (pDMN +vDMN) where pDMN_to_vDMN and pDMN_to_adDMN are correlations between time courses in the posterior DMN, ventral DMN and anterior‐dorsal DMN. pDMN and vDMN are the within‐region correlations. NFQ values from 1447 scans of 739 individual study participants (Table 1) were analyzed using linear mixed effects models to account for age, clinical diagnosis, vendor, and time trends.ResultBox plots of NFQ values by diagnosis within ADNI‐phase and protocol are shown in Figure 2 and suggest important differences among clinical diagnoses and between ADNI‐2 and ADNI‐3. The mixed model analysis indicates that a 10‐year increase in age is associated with a 0.05±0.01 (mean±SD) increase in NFQ; individuals with MCI average roughly 0.05±0.01 higher than unimpaired and individuals with dementia average 0.14±0.02 higher. Differences by vendor for ADNI‐2 and ADNI‐3 are summarized in Figures 2 and 3 with effect sizes on par with these clinical factors. Mean NFQ for Phillips ADNI‐2 were highest (0.54±0.02); somewhat lower were Philips ADNI‐3 (0.51±0.02) and Siemens ADNI‐3 advanced (0.49±0.03); Siemens ADNI‐3 basic (0.40±0.02) and GE ADNI‐3 were the lowest (0.38±0.02).ConclusionNFQ values consistently increase with impairment. Vendor‐specific differences are observed and vendor changes over the life of a study introduce variability on the scale of disease‐related changes in NFQ.

AbstractBackgroundThe NIA‐AA Research Framework for Alzheimer’s disease (AD) includes consideration of two clinical schemes for staging AD: (1) traditional cognitively unimpaired (CU) to MCI to dementia and (2) a new numeric staging (1‐6) for individuals who are on the Alzheimer’s continuum.MethodThe Mayo Clinic Study of Aging is a population‐based longitudinal study of aging and cognitive impairment in Olmsted County, Minnesota. We analyzed CU persons who had completed three consecutive clinical visits and calculated measures for cross‐sectional categories: objective cognitive impairment and function and measures for change including subjective cognitive impairment (SCD), objective cognitive change (ΔOBJ), and new onset of neurobehavioral symptoms (NBS). We calculated the frequencies for Stage 2 by varying the cutoffs for each of the measures above and assessed stability of this stage over 15 months.ResultThe frequency of Stage 2 varied from 5‐48% depending upon the age of the participants and the cutoff scores used. Membership in Stage 2 required abnormal SCD, ΔOBJ, or ΔNBS and having only ΔOBJ abnormal was the primary reason for inclusion in Stage 2 (45‐62%) followed by 10‐21% having only SCD abnormal. The stability of Stage 2 over 15 months ranged from 31‐65% with 5‐13% progressing to Stage 3 or higher and 22‐43% percent reverting to Stage 1.ConclusionStage 2 was characterized as a transitional period between CU and MCI and could be operationalized with a variety of measures. The staging scheme performed well considering the variables evaluated and cutoff scores employed.

Summary This paper provides guidance for researchers with some mathematical background on the conduct of time‐to‐event analysis in observational studies based on intensity (hazard) models. Discussions of basic concepts like time axis, event definition and censoring are given. Hazard models are introduced, with special emphasis on the Cox proportional hazards regression model. We provide check lists that may be useful both when fitting the model and assessing its goodness of fit and when interpreting the results. Special attention is paid to how to avoid problems with immortal time bias by introducing time‐dependent covariates. We discuss prediction based on hazard models and difficulties when attempting to draw proper causal conclusions from such models. Finally, we present a series of examples where the methods and check lists are exemplified. Computational details and implementation using the freely available R software are documented in Supplementary Material. The paper was prepared as part of the STRATOS initiative.

Abstract Clinical trials with anti-tau drugs will need to target individuals at risk of accumulating tau. Our objective was to identify variables available in a research setting that predict future rates of tau PET accumulation separately among individuals who were either cognitively unimpaired or cognitively impaired. All 337 participants had: a baseline study visit with MRI, amyloid PET, and tau PET exams, at least one follow-up tau PET exam; and met clinical criteria for membership in one of two clinical diagnostic groups: cognitively unimpaired (n = 203); or cognitively impaired (n = 134, a combined group of participants with either mild cognitive impairment or dementia with Alzheimer’s clinical syndrome). Our primary analyses were in these two clinical groups; however, we also evaluated subgroups dividing the unimpaired group by normal/abnormal amyloid PET and the impaired group by clinical phenotype (mild cognitive impairment, amnestic dementia, and non-amnestic dementia). Linear mixed effects models were used to estimate associations between age, sex, education, APOE genotype, amyloid and tau PET standardized uptake value ratio (SUVR), cognitive performance, cortical thickness, and white matter hyperintensity volume at baseline, and the rate of subsequent tau PET accumulation. Log-transformed tau PET SUVR was used as the response and rates were summarized as annual per cent change. A temporal lobe tau PET meta-region of interest was used. In the cognitively unimpaired group, only higher baseline amyloid PET was a significant independent predictor of higher tau accumulation rates (P < 0.001). Higher rates of tau accumulation were associated with faster rates of cognitive decline in the cognitively unimpaired subgroup with abnormal amyloid PET (P = 0.03), but among the subgroup with normal amyloid PET. In the cognitively impaired group, younger age (P = 0.02), higher baseline amyloid PET (P = 0.05), APOE ε4 (P = 0.05), and better cognitive performance (P = 0.05) were significant independent predictors of higher tau accumulation rates. Among impaired individuals, faster cognitive decline was associated with faster rates of tau accumulation (P = 0.01). While we examined many possible predictor variables, our results indicate that screening of unimpaired individuals for potential inclusion in anti-tau trials may be straightforward because the only independent predictor of high tau rates was amyloidosis. In cognitively impaired individuals, imaging and clinical variables consistent with early onset Alzheimer’s disease phenotype were associated with higher rates of tau PET accumulation suggesting this may be a highly advantageous group in which to conduct proof-of-concept clinical trials that target tau-related mechanisms. The nature of the dementia phenotype (amnestic versus non-amnestic) did not affect this conclusion.

Abstract Background: Monoclonal gammopathy of undetermined significance (MGUS) is a premalignant plasma cell disorder that carries a 1% per year risk of progression to multiple myeloma (MM) or related malignancy. The prevalence and natural history of MGUS, in which by definition intact immunoglobulin heavy chain (IgH) is expressed, has been well described. However, up to 20% of myeloma (MM) is characterized by complete lack of IgH expression (Light-chain MM); the prevalence of a corresponding precursor entity, light chain MGUS (LC-MGUS) has not been determined. We report the first prevalence estimates of LC-MGUS in the general population from a large, well-defined geographic population using modern laboratory techniques. Methods: The cohort for this study was derived from one previously assembled by us to estimate the prevalence of MGUS (N Engl J Med2006;354:1362-9). The original cohort used to estimate the prevalence of MGUS consisted of 21,463 of the 28,038 enumerated residents aged 50 or over of Olmsted County Minnesota as of January 1, 1995. The sensitive serum free light chain (FLC) assay (The Binding Site Limited, Birmingham, U.K.) was performed on stored serum samples from these 21,463 persons. IgH expression was determined by immunofixation on all FLC results that had an abnormal kappa/lambda ratio (<0.26 or >1.65). LC-MGUS was defined as the presence of an abnormal FLC ratio and a negative immunofixation for IgH expression. Results: Adequate stored serum samples were available in 20,733 (97%) of the 21,463 persons. To date, the FLC assay has been performed and results were available for analysis on samples from 16,637 persons. An abnormal FLC ratio was observed in 572 persons. IgH expression was detected in 255 of these cases on immunofixation; these persons are considered as having MGUS, and were excluded from the estimation of LC-MGUS prevalence. This resulted in 317 persons out of 16,637 who had an abnormal FLC ratio without evidence of IgH expression, resulting in an estimated prevalence of LC- MGUS of 2%. Of the 317 cases of LC-MGUS identified in this study, 217 were kappa and 100 were lambda; in 35 cases the presence of the corresponding monoclonal light chain was apparent on immunofixation. The median age of the cohort of LC-MGUS was 62 years; males=151, females =166. The involved FLC level ranged from 0.118–270.0 mg/dL. The FLC ratio ranged from 0.014–0.253 (lambda) and 1.67–511.01 (kappa). So far, progression to multiple myeloma has occurred in 4 patients, a rate much higher than what is expected based on the prevalence of myeloma in the general population. Two additional patients have developed CLL. Conclusions: LC-MGUS is prevalent in 2% of the general population aged 50 years of age or older. The natural history of this disorder needs to be determined.

Abstract Smoldering multiple myeloma (SMM) is characterized by a serum M protein ≥ 3g/dL and/or 10% or more of plasma cells in the bone marrow. However, the definition is not standardized, and it is not known whether both serum M protein levels and bone marrow plasma cell counts are necessary for diagnosis or if one parameter is sufficient. We reviewed the medical records and bone marrows of all patients from Mayo Clinic seen within 30 days of recognition of an IgG or IgA M protein ≥ 3g/dL or a bone marrow containing ≥ 10% plasma cells from 1970 to 1995. This allows for a minimum potential follow-up of 10 years. Patients with end-organ damage at baseline from plasma cell proliferation, including active multiple myeloma (MM) and primary amyloidosis (AL) and those who had received chemotherapy were excluded. A differential of the bone marrow aspirate coupled with the bone marrow biopsy morphology and immunohistochemistry using antibodies directed against CD138, MUM-1 and Cyclin D1 were evaluated in every case in order to estimate the plasma cell content. In all, 301 patients fulfilled either of the criteria for SMM. Their median age was 64 years and only 3% were less than 40 years of age; 60% were male. The median hemoglobin value was 12.9 g/dL; 7% were less than 10 g/dL, but the anemia was unrelated to plasma cell proliferation. IgG accounted for 75%, IgA 22%, and biclonal proteins were found in 3%. The serum light-chain was κ in 67% and λ in 33%. The median serum M spike was 2.9 g/dL; 11% were at least 4.0 g/dL. Uninvolved serum immunoglobulins were reduced in 81%; only 1 immunoglobulin was reduced in 31% and both were decreased in 50%. The urine contained a monoclonal κ protein in 36% and λ in 18% and 46% were negative. The median size of the urine M spike was 0.04 g/24h; only 5 (3%) were > 1 g/24h. The median bone marrow plasma cell content was 15 – 19%; 10% had less than 10% plasma cells, while 10% had at least 50% plasma cells in the bone marrow. Cyclin D-1 was expressed in 17%. Patients were categorized into 3 groups: Group 1, serum M protein ≥ 3g/dL and bone marrow containing ≥ 10% plasma cells (n= 113, 38%); Group 2, bone marrow plasma cells ≥ 10% but serum M protein < 3g/dL (n= 158, 52%); Group 3, serum M protein ≥ 3g/dL but bone marrow plasma cells < 10% (n= 30, 10%). During 2,204 cumulative years of follow-up 85% died (median follow-up of those still living 10.8 years), 155 (51%) developed MM, while 7 (2%) developed AL. The overall rate of progression at 10 years was 62%; median time to progression was 5.5 yrs. The median time to progression was 2.4, 9.2, and 19 years in groups 1, 2, and 3 respectively; correspondingly at 10 years, progression occurred in 76%, 59%, and 32% respectively. Significant risk factors for progression with univariate analysis were serum M spike ≥ 4g/dL (p < 0.001), presence of IgA (p = 0.003), presence of urine light chain (p = 0.006), presence of λ urinary light chain (p = 0.002), bone marrow plasma cells ≥ 20% (p < 0.001) and reduction of uninvolved immunoglobulins (p < 0.001). The hemoglobin value, gender, serum albumin, and expression of cyclin D-1 were not of prognostic importance. On multivariate analysis, the percentage of bone marrow plasma cells was the only significant factor predicting progression to MM or AL.

Abstract Background: Solitary plasmacytoma of bone (SBP) is a localized collection of monoclonal plasma cells that is potentially curable with local radiation therapy but associated with a high risk of progression to multiple myeloma. We hypothesized that an abnormal immunoglobulin free light (FLC) ratio at diagnosis may be a prognostic indicator of transformation risk. Methods: We identified a cohort of 133 patients with SBP for whom stored serum taken at the time of diagnosis was available. The diagnosis was ascertained and serum FLC determined in 126 patients. Results: From this cohort, 48 patients have progressed to myeloma and the median time to progression among those who progressed was 1.9 years. On univariate analysis, age (p<0.001), gender (p=0.035), abnormal FLC ratio at diagnosis (p=0.009) and persistence of serum or urine M-protein after therapy (p=0.0070 were all associated with a shorter overall survival (OS) and time to progression to multiple myeloma. Progression by Normal FLC(0.26–1.65) Progression by Normal FLC(0.26–1.65) On multivariate analysis, an abnormal FLC ratio retained its independence in a model that includes age at diagnosis but lost its significance when combined with persistence of the serum or urine M-protein. However, serum or urine M-proteins are not detectable in a significant number of patients with SBP and therefore not informative. Conclusion: The FLC ratio at the time of diagnosis of SBP is a powerful predictor of risk and a useful aid to management of patients with this condition.

Abstract Background: Monoclonal gammopathy of undetermined significance (MGUS) is a common pre-malignant plasma cell disorder associated with a 1% per year risk of progression to multiple myeloma or related malignancy. Since the risk factors for MGUS are poorly defined, the goal of the present study was to determine if the risk of MGUS is increased in first degree relatives of patients with known multiple myeloma (MM) or MGUS. Methods: MGUS probands (index cases) were recruited from a population-based prevalence study of MGUS in Olmsted County, MN while MM probands were recruited through the Mayo Clinic practice. Consenting probands were asked to provide contact information on all first-degree relatives ages 40 years and older. Serum samples were then collected from first-degree relatives with informed consent and subjected to agaraose-gel electrophoresis and immunofixation. The prevalence of MGUS in first-degree relatives of MM and MGUS probands was compared to population-based rates from Olmsted County using risk ratios (RR). For comparisons to the reference population, only cases detected by protein electrophoresis and confirmed by immunofixation were included so that the diagnostic strategy was identical in the two groups being compared. Results: Serum samples were obtained from 911 relatives of 329 unique families, including 493 siblings, 324 children and 94 parents of patients with MGUS or MM. Using protein electrophoresis, monoclonal protein was detected in the serum of 55 (6%) while immunofixation identified 28 additional relatives (3%), for an age- and sex-adjusted prevalence rate of 8.1% (95%CI: 6.3, 9.8). The age-specific prevalence of MGUS in first-degree relatives increased with age (1.9%, 6.9%, 11.6%, 14.6%, and 21.0% for ages 40–49, 50–59, 60–69, 70–79 and ≥80, respectively; P<0.001). Based on similar MGUS detection methodology as the reference population, there was a significantly higher risk of MGUS in first-degree relatives (age-adjusted RR, 2.6, 95%CI: 1.9, 3.4) compared to the reference population. The increased risk (P<0.001) was seen both in relatives of MM probands (RR, 2.0, 95%CI: 1.4, 2.8), as well as MGUS probands (RR, 3.3, 95%CI: 2.1, 4.8). This association was similar across age of proband, age and gender of relative, and relationship of the first-degree relative. When examining whether the increased risk of MGUS in relatives was specific to probands with a large monoclonal protein concentration or specific monoclonal immunoglobulin isotype (i.e. high-risk MGUS phenotype), there was suggestion of a greater risk for relatives of probands with a high (3 1.5 g/dL) M-protein (RR, 2.8, 95%CI: 2.0,3.8) compared to lower M-protein levels (RR, 1.8, 95%CI: 1.1,2.8) although the difference did not reach statistical significance (P=0.12). The risk of MGUS in relatives did not differ by proband’s isotype (IgG vs. other). Conclusions: First-degree relatives of patients with MM or MGUS have a greater than two-fold risk of MGUS compared to the general population, implying underlying genetic predisposition for these diseases and providing rationale for identifying genetic determinants of MGUS. This study also provides important baseline rates of MGUS in first-degree relatives that impact the clinical care of these patients.