AbstractObjective:To identify peri-conceptional diet patterns among women in Bangalore and examine their associations with risk of gestational diabetes mellitus (GDM).Design:BAngalore Nutrition Gestational diabetes LifEstyle Study, started in June 2016, was a prospective observational study, in which women were recruited at 5–16 weeks’ gestation. Peri-conceptional diet was recalled at recruitment, using a validated 224-item FFQ. GDM was assessed by a 75-g oral glucose tolerance test at 24–28 weeks’ gestation, applying WHO 2013 criteria. Diet patterns were identified using principal component analysis, and diet pattern–GDM associations were examined using multivariate logistic regression, adjusting for ‘a priori’ confounders.Setting:Antenatal clinics of two hospitals, Bangalore, South India.Participants:Seven hundred and eighty-five pregnant women of varied socio-economic status.Results:GDM prevalence was 22 %. Three diet patterns were identified: (a) high-diversity, urban (HDU) characterised by diverse, home-cooked and processed foods was associated with older, more affluent, better-educated and urban women; (b) rice-fried snacks-chicken-sweets (RFCS), characterised by low diet diversity, was associated with younger, less-educated, and lower-income, rural and joint families; and (c) healthy, traditional vegetarian (HTV), characterised by home-cooked vegetarian and non-processed foods, was associated with less-educated, more affluent, and rural and joint families. The HDU pattern was associated with a lower GDM risk (adjusted odds ratio (aOR): 0·80/sd, 95 % CI (0·64, 0·99), P = 0·04) after adjusting for confounders. BMI was strongly related to GDM risk and possibly mediated diet–GDM associations.Conclusions:The findings support global recommendations to encourage women to attain a healthy pre-pregnancy BMI and increase diet diversity. Both healthy and unhealthy foods in the patterns indicate low awareness about healthy foods and a need for public education.

Automatic differentiation (AD) is conventionally understood as a family of distinct algorithms, rooted in two "modes"-forward and reverse-which are typically presented (and implemented) separately. Can there be only one? Following up on the AD systems developed in the JAX and Dex projects, we formalize a decomposition of reverse-mode AD into (i) forward-mode AD followed by (ii) unzipping the linear and non-linear parts and then (iii) transposition of the linear part. To that end, we define a (substructurally) linear type system that can prove a class of functions are (algebraically) linear. Our main results are that forward-mode AD produces such linear functions, and that we can unzip and transpose any such linear function, conserving cost, size, and linearity. Composing these three transformations recovers reverse-mode AD. This decomposition also sheds light on checkpointing, which emerges naturally from a free choice in unzipping let expressions. As a corollary, checkpointing techniques are applicable to general-purpose partial evaluation, not just AD. We hope that our formalization will lead to a deeper understanding of automatic differentiation and that it will simplify implementations, by separating the concerns of differentiation proper from the concerns of gaining efficiency (namely, separating the derivative computation from the act of running it backward).

Introduction & Objectives Idiopathic arterial pulmonary hypertension (IPAH) is a chronic, progressive respiratory disease, characterised by elevated pulmonary artery pressure. The disease carries significant mortality and therefore, emphasis is placed on identification of accurate outcome predictors. The Distance saturation Product (DSP) is a novel index that has demonstrated prognostic value in cardio-respiratory diseases. The 'six-minute walk test' (6MWT) is a submaximal-effort exercise test used in diagnosis and ongoing management of patients with IPAH. The DSP is the product of the distance walked and lowest oxygen saturation recorded during the six-minute walk test. We aimed to evaluate if the DSP could: a) predict outcomes in IPAH; and b) correlate with other clinical parameters. Methods A retrospective analysis was performed of 146 patients with IPAH attending the pulmonary vascular unit between July 2011 and May 2021. Receiver operating characteristic (ROC) and kaplan meier curves evaluated the ability of the DSP, 6-minute walk distance (6MWD) and DDSP to predict patient mortality. Pearson's correlation coefficient assessed the correlation of the DSP with other clinical parameters. Results Of the 146 patients analysed, 55(37.7%) were dead at the point of censorship. The mean DSP, 6MWD and DDSP were 224.2m% ± 140.5, 258.1m ± 147.1 and 43.59m% ± 88.67 respectively. Baseline DSP was the strongest predictor of mortality in ROC analysis (AUC=0.7364, p≤0.0001), followed by baseline 6MWD (AUC=0.7198, p≤0.0001) and, lastly, DDSP (AUC=0.5892, p=0.1596). A baseline DSP >181.8m% was associated with 4.4 times increased risk of mortality. The strongest correlation was between DSP and NTproBNP (r=-0.3271, p≤0.0001), followed by Emphasis10 score (r=-0.3851, p≤0.0001) and lastly mixed venous oxygen saturation (r=-0.3571, p≤0.0001). Conclusion DSP represents a simple tool for predicting survival of patients with IPAH. Patients with a baseline DSP <181.8m% are at significantly increased risk of death. The DSP demonstrated moderate correlations with NTproBNP measurements, Emphasis10 scores and SaO2 levels. The DSP's potential utility in the evaluation of IPAH patients is supported by its simplicity, ease of calculation and accessibility of the 6MWT in clinical settings. To support these preliminary findings, further study is necessary to validate the DSP as an outcome predictor in IPAH.

Introducción: Las enfermedades relacionadas al estilo de vida son uno de los mayores retos de salud del siglo 21. Objetivos: El propósito de esta investigación fue obtener una base de datos para estudiar la prevalencia de enfermedades de las personas que viven en pobreza en Lima, Perú. Metodología: La investigación estuvo localizada en los distritos de Comas y Carabayllo en Lima, Perú. Contamos con un total de 829 adultos y 770 niños (0-17 años de edad) participantes. La data fue recolectada a través de clínicas comunitarias gratuitas, estas incluyeron muestras de sangre para evaluar la hemoglobina, glucosa, hemoglobina glicosilada, lípidos, vitamina D, y anticuerpos en contra de Chagas y Helicobacter pylori. Para la población pediátrica sólo se utilizó los records médicos; no se utilizaron muestras de sangre con propósitos de investigación. Resultados: Los resultados más significativos fueron: 50,9% con presión arterial sanguínea elevada siendo sistólica o diastólica, 47% Con hemoglobina glicosilada elevada, 24% glucosa en ayuno elevada, 57,2% con un al menos un parámetro elevado del panel lípido, 32,6% hemoglobina baja, 97,2% Vitamina D baja, 59% positivo para anticuerpos de Helicobacter, y 5,6% positivo con anticuerpos de Chagas. La prevalencia de sobrepeso y obesidad fue 65,1% para adultos y 42,3% para la población pediátrica. Conclusión: Los resultados demuestran anomalías relacionadas al estilo de vida. Esta información puede utilizarse para desarrollar estrategias de prevención y tratamiento de las enfermedades relacionadas al estilo de vida, con enfoque en la educación y cambios en el estilo de vida. DOI: https://doi.org/10.25176/RFMH.v17.n2.830

Social influences may create a barrier to couples HIV testing and counselling (CHTC) uptake in sub-Saharan Africa. This secondary analysis of data collected in the ‘Uthando Lwethu’ randomised controlled trial used discrete-time survival models to evaluate the association between within-couple average ‘peer support’ score and uptake of CHTC by the end of nine months’ follow-up. Peer support was conceptualised by self-rated strength of agreement with two statements describing friendships outside of the primary partnership. Eighty-eight couples (26.9%) took up CHTC. Results tended towards a dichotomous trend in models adjusted only for trial arm, with uptake significantly less likely amongst couples in the higher of four peer support score categories (OR 0.34, 95% CI 0.18, 0.68 [7–10 points]; OR 0.53, 95% CI 0.28, 0.99 [≥ 11 points]). A similar trend remained in the final multivariable model, but was no longer significant (AOR 0.59, 95% CI 0.25, 1.42 [7–10 points]; AOR 0.88, 95% CI 0.36, 2.10 [≥ 11 points]). Accounting for social influences in the design of couples-focused interventions may increase their success.

We present a novel programming language design that attempts to combine the clarity and safety of high-level functional languages with the efficiency and parallelism of low-level numerical languages. We treat arrays as eagerly-memoized functions on typed index sets, allowing abstract function manipulations, such as currying, to work on arrays. In contrast to composing primitive bulk-array operations, we argue for an explicit nested indexing style that mirrors application of functions to arguments. We also introduce a fine-grained typed effects system which affords concise and automatically-parallelized in-place updates. Specifically, an associative accumulation effect allows reverse-mode automatic differentiation of in-place updates in a way that preserves parallelism. Empirically, we benchmark against the Futhark array programming language, and demonstrate that aggressive inlining and type-driven compilation allows array programs to be written in an expressive, pointful style with little performance penalty.

COPD remains largely undiagnosed or is diagnosed late in the course of disease. We report findings of a specialist outreach programme to identify undiagnosed COPD in primary care. An electronic case-finding algorithm identified 1602 at-risk patients from 12 practices who were invited to attend the clinic. Three hundred and eighty-three (23.9%) responded and 288 were enrolled into the study. Forty-eight (16.6%) had undiagnosed mild and 28 (9.7%) had moderate airway obstruction, meeting spirometric diagnostic criteria for COPD. However, at 12 months only 8 suspected COPD patients (10.6%) had received a diagnostic label in their primary care record. This constituted 0.38% of the total patient population, as compared with 0.31% of control practices, p = 0.306. However, if all patients with airway obstruction received a coding of COPD, then the diagnosis rate in the intervention group would have risen by 0.84%. Despite the low take-up and diagnostic yield, this programme suggests that integrated case-finding strategies could improve COPD recognition.

We decompose reverse-mode automatic differentiation into (forward-mode) linearization followed by transposition. Doing so isolates the essential difference between forward- and reverse-mode AD, and simplifies their joint implementation. In particular, once forward-mode AD rules are defined for every primitive operation in a source language, only linear primitives require an additional transposition rule in order to arrive at a complete reverse-mode AD implementation. This is how reverse-mode AD is written in JAX and Dex.

We develop automatic differentiation (AD) procedures for reductions and scans—parameterized by arbitrary differentiable monoids—in a way that preserves parallelism, by rewriting them as other reductions and scans. This is in contrast with the literature and with existing AD systems, which are either general, but force sequential execution of the derivative program, or only include hand-crafted rules for a select few monoids (usually (0, +), (1, ×), (−∞, max) and (∞, min)) and thus lack the general flexibility of second-order languages.

To estimate the prevalence, incidence and predictors of cardiovascular disease (CVD) risk factors in the Vellore Birth Cohort, South India.Prospective, cohort studyPopulation-based cohort of rural and urban communities in and around Vellore city in South IndiaNon-migrant individuals (n= 962, male 519) were studied at two time points 13.6 years apart i) 1998-2002 (baseline, mean age 28.2 years) and ii) 2013-2014 (follow-up, mean age 41.7 years).Prevalence and incidence of CVD risk factors (obesity, central obesity, type 2 diabetes (T2D), hypertension, hypercholesterolemia and hypertriglyceridemia) studied at baseline (1998-2002) and follow-up (2013-2014), prevalence in comparison with the Non-Communicable Disease Risk Collaboration (global) data, incidence in comparison with another Indian cohort from New Delhi (NDBC), and baseline predictors of incident CVD risk factors.The prevalence at 28 and 42 years was 17% and 51% for overweight/obesity, 19% and 59% for central obesity, 3% and 16% for T2D, 2% and 19% for hypertension and 15% and 30% for hypertriglyceridemia. The prevalence of T2D at baseline and follow-up and hypertension at follow-up was comparable with or exceeded that in high income countries despite lower obesity rates. The incidence of most risk factors was lower in Vellore than in the NDBC. Waist circumference strongly predicted incident T2D, hypertension and hypertriglyceridemia.A high prevalence of CVD risk factors was evident at a young age among Indians compared with high and upper-middle income countries, with rural rates catching up with urban estimates. Adiposity predicted higher incident CVD risk, but the prevalence of hypertension and T2D was higher given a relatively low obesity prevalence in global terms. Our findings highlight a high burden of CVD risk factors at younger age with increasing trends observed among rural residents, similar to urban South Indians. Therefore, strategies to prevent CVD should be strengthened in both rural and urban settings to minimise health inequalities and should start young.NoneCardiovascular disease (CVD) risk burden is increasing in Low- and middle-income countries and contributes significantly to the overall morbidity and mortality.Nation-wide data from India demonstrate heterogeneity in the prevalence of CVD risk factors within the country; there is very little incidence data.The prevalence of CVD risk factors in India is comparable with or exceeds that in high income countries like USA and Europe, even though obesity levels are lower.Adiposity at baseline, particularly waist circumference, is a strong predictor of incident risk factors.The prevalence of CVD risk factors is higher in rural than urban communities, but the incidence is comparable or higher in the rural setting indicating that the rural population are catching up

Background. Acute kidney injury (AKI) is common and is associated with significant morbidity and mortality. Socioeconomic status may be negatively associated with AKI as some risk factors for AKI such as chronic kidney disease, diabetes and heart failure are socially distributed. This study explored the socioeconomic gradient of the incidence and mortality of AKI, after adjusting for important mediators such as comorbidities. Methods. Linked primary care and laboratory data from two large acute hospitals in the south of England, sourced from the Care and Health Information Analytics database, were used to identify AKI cases over a 1-year period (2017-18) from a population of 580 940 adults. AKI was diagnosed from serum creatinine patterns using a Kidney Disease: Improving Global Outcomes-based definition. Multivariable logistic regression and Cox proportional hazard models adjusting for age, sex, comorbidities and prescribed medication (in incidence analyses) and AKI severity (in mortality analyses), were used to assess the association of area deprivation (using Index of Multiple Deprivation for place of residence) with AKI risk and all-cause mortality over a median (interquartile range) of 234 days (119-356). Results. Annual incidence rate of first AKI was 1726/100 000 (1.7%). The risk of AKI was higher in the most deprived compared with the least deprived areas [adjusted odds ratio = 1.79, 95% confidence interval (CI) 1.59-2.01 and 1.33, 95% CI 1.03-1.72 for <65 and >65 year old, respectively] after controlling for age, sex, comorbidities and prescribed medication. Adjusted risk of mortality post first AKI was higher in the most deprived areas (adjusted hazard ratio = 1.20, 95% CI 1.07-1.36). Conclusions. Social deprivation was associated with higher incidence of AKI and poorer survival even after adjusting for the higher presence of comorbidities. Such social inequity should be considered when devising strategies to prevent AKI and improve care for AKI patients.

Understanding how genes, drugs and neural circuits influence behavior requires the ability to effectively organize information about similarities and differences within complex behavioral datasets. Motion Sequencing (MoSeq) is an ethologically inspired behavioral analysis method that identifies modular components of three-dimensional mouse body language called 'syllables'. Here, we show that MoSeq effectively parses behavioral differences and captures similarities elicited by a panel of neuroactive and psychoactive drugs administered to a cohort of nearly 700 mice. MoSeq identifies syllables that are characteristic of individual drugs, a finding we leverage to reveal specific on- and off-target effects of both established and candidate therapeutics in a mouse model of autism spectrum disorder. These results demonstrate that MoSeq can meaningfully organize large-scale behavioral data, illustrate the power of a fundamentally modular description of behavior and suggest that behavioral syllables represent a new class of druggable target. By analyzing hundreds of mice treated with a library of neuro- and psychoactive drugs, Wiltschko et al. show that Motion Sequencing can effectively discriminate and categorize drug effects and link molecular targets to behavioral syllables.

Populations of wild spring-run Chinook salmon in California’s Central Valley, once numbering in the millions, have dramatically declined to record low numbers. Dam construction, habitat degradation, and altered flow regimes have all contributed to depress populations, which currently persist in only a few tributaries to the Sacramento River. Mill Creek (Tehama County) continues to support these threatened fish, and contains some of the most pristine spawning and rearing habitat available in the Central Valley. Despite this pristine habitat, the number of Chinook salmon returning to spawn has declined to record low numbers, likely due to poor outmigration survival rates. From 2013 to 2017, 334 smolts were captured and acoustic tagged while out-migrating from Mill Creek, allowing for movement and survival rates to be tracked over 250 km through the Sacramento River. During this study California experienced both a historic drought and record rainfall, resulting in dramatic fluctuations in year-to-year river flow and water temperature. Cumulative survival of tagged smolts from Mill Creek through the Sacramento River was 9.5% (±1.6) during the study, with relatively low survival during historic drought conditions in 2015 (4.9% ± 1.6) followed by increased survival during high flows in 2017 (42.3% ± 9.1). Survival in Mill Creek and the Sacramento River was modeled over a range of flow values, which indicated that higher flows in each region result in increased survival rates. Survival estimates gathered in this study can help focus management and restoration actions over a relatively long migration corridor to specific regions of low survival, and provide guidance for management actions in the Sacramento River aimed at restoring populations of threatened Central Valley spring-run Chinook salmon.

Differential equations parameterized by neural networks become expensive to solve numerically as training progresses. We propose a remedy that encourages learned dynamics to be easier to solve. Specifically, we introduce a differentiable surrogate for the time cost of standard numerical solvers, using higher-order derivatives of solution trajectories. These derivatives are efficient to compute with Taylor-mode automatic differentiation. Optimizing this additional objective trades model performance against the time cost of solving the learned dynamics. We demonstrate our approach by training substantially faster, while nearly as accurate, models in supervised classification, density estimation, and time-series modelling tasks.

IntroductionIndia has high mortality rates from cardiovascular disease (CVD). Understanding the trends and identifying modifiable determinants of CVD risk factors will guide preventive strategies and policy making.Research design and methodsCVD risk factors (obesity, central obesity, and type 2 diabetes (T2D), hypertension, hypercholesterolemia and hypertriglyceridemia) prevalence and incidence were estimated in 962 (male 519) non-migrant adults from Vellore, South India, studied in: (1) 1998–2002 (mean age 28.2 years) and (2) 2013–2014 (mean age 41.7 years). Prevalence was compared with the Non-Communicable Disease Risk Collaboration (global) data. Incidence was compared with another Indian cohort from New Delhi Birth Cohort (NDBC). Regression analysis was used to test baseline predictors of incident CVD risk factors.ResultsThe prevalence at 28 and 42 years was 17% (95% CI 14% to 19%) and 51% (95% CI 48% to 55%) for overweight/obesity, 19% (95% CI 17% to 22%) and 59% (95% CI 56% to 62%) for central obesity, 3% (95% CI 2% to 4%) and 16% (95% CI 14% to 19%) for T2D, 2% (95% CI 1% to 3%) and 19% (95% CI 17% to 22%) for hypertension and 15% (95% CI 13% to 18%) and 30% (95% CI 27% to 33%) for hypertriglyceridemia. The prevalence of T2D at baseline and follow-up and hypertension at follow-up was comparable with or exceeded that in high-income countries despite lower obesity rates. The incidence of most risk factors was lower in Vellore than in the NDBC. Waist circumference strongly predicted incident T2D, hypertension and hypertriglyceridemia.ConclusionsA high prevalence of CVD risk factors was evident at a young age among Indians compared with high and upper middle income countries, with rural rates catching up with urban estimates. Adiposity predicted higher incident CVD risk, but the prevalence of hypertension and T2D was higher given a relatively low obesity prevalence. Preventive efforts should target both rural and urban India and should start young.

In variational autoencoders, the prior on the latent codes $z$ is often treated as an afterthought, but the prior shapes the kind of latent representation that the model learns. If the goal is to learn a representation that is interpretable and useful, then the prior should reflect the ways in which the high-level factors that describe the data vary. The "default" prior is an isotropic normal, but if the natural factors of variation in the dataset exhibit discrete structure or are not independent, then the isotropic-normal prior will actually encourage learning representations that mask this structure. To alleviate this problem, we propose using a flexible Bayesian nonparametric hierarchical clustering prior based on the time-marginalized coalescent (TMC). To scale learning to large datasets, we develop a new inducing-point approximation and inference algorithm. We then apply the method without supervision to several datasets and examine the interpretability and practical performance of the inferred hierarchies and learned latent space.

Most clinical contacts with chronic obstructive pulmonary disease (COPD) patients take place in primary care, presenting opportunity for proactive clinical management. Electronic health records could be used to risk stratify diagnosed patients in this setting, but may be limited by poor data quality or completeness. We developed a risk stratification database algorithm using the DOSE index (Dyspnoea, Obstruction, Smoking and Exacerbation) with routinely collected primary care data, aiming to calculate up to three repeated risk scores per patient over five years, each separated by at least one year. Among 10,393 patients with diagnosed COPD, sufficient primary care data were present to calculate at least one risk score for 77.4%, and the maximum of three risk scores for 50.6%. Linked secondary care data revealed primary care under-recording of hospital exacerbations, which translated to a slight, non-significant cohort average risk score reduction, and an understated risk group allocation for less than 1% of patients. Algorithmic calculation of the DOSE index is possible using primary care data, and appears robust to the absence of linked secondary care data, if unavailable. The DOSE index appears a simple and practical means of incorporating risk stratification into the routine primary care of COPD patients, but further research is needed to evaluate its clinical utility in this setting. Although secondary analysis of routinely collected primary care data could benefit clinicians, patients and the health system, standardised data collection and improved data quality and completeness are also needed.

Background The deep venous thrombosis (DVT) prevalence in advanced cancer is unconfirmed and it is unknown whether current international thromboprophylaxis guidance is applicable to this group. We determined prevalence and predictors of femoral DVT in patients admitted to specialist palliative care units (SPCU). Methods Prospective longitudinal observational study in five SPCUs in England, Wales and Northern Ireland. Consecutive adults with cancer underwent bilateral femoral vein ultrasonography on admission and weekly until death or discharger for a maximum of three weeks. Data were collected on performance status, attributable symptoms and variable known to be associated with venous thromboembolism. Patients were ineligible if admitted for terminal care (estimated prognosis <five days). Prevalence was estimated with 95% confidence intervals (CI). DVT predictors and survival were explored using regression analyses. Sensitivity analysis excluded early scans to account for a technical learning curve. Results 343 participants (68·2 [SD 12·8] 25 to 102 years; men 52%; AKPS 49 [SD 16·6] 20% to 90%) were recruited. Of 273 evaluable scans, 92 (34%, CI 28% to 40%) showed DVT. Excluding early scans, 64/232 (28%, 22% to 34%) showed DVT. Four participants with a ‘no DVT’ scan on admission developed a DVT on repeat scanning over 21 days. Previous thromboembolism, bedbound ≤12 weeks for any reason (p=0.003) and lower limb oedema (p=0.009) independently predicted DVT. Serum albumin (p=0.430), thromboprophylaxis (p=0.173) and survival (p=0.473) were unrelated to DVT. Conclusions These novel data show approximately one third of SPCU admissions with advanced cancer had a femoral DVT. DVT was not associated with thromboprophylaxis, survival or symptoms other than leg oedema. Findings are consistent with VTE being a manifestation of advanced disease rather than a cause of premature death. Thromboprophylaxis for SPCU in-patients with poor performance status seems of little benefit.

Chronic obstructive pulmonary disease (COPD) is heterogeneous, but persistent airflow obstruction (AFO) is fundamental to diagnosis. We studied AFO consistency from initial diagnosis and explored factors associated with absent or inconsistent AFO. This was a retrospective observational study using patient-anonymised routine individual data in Care and Health Information Analytics (CHIA) database. Identifying a prevalent COPD cohort based on diagnostic codes in primary care records, we used serial ratios of forced expiratory volume in 1 s to forced vital capacity (FEV1/FVC%) from time of initial COPD diagnosis to assign patients to one of three AFO categories, according to whether all (persistent), some (variable) or none (absent) were <70%. We described respiratory prescriptions over 3 years (2011–2013) and used multivariable logistic regression to estimate odds of absent or variable AFO and potential predictors. We identified 14,378 patients with diagnosed COPD (mean ± SD age 68.8 ± 10.7 years), median (IQR) COPD duration of 60 (25,103) months. FEV1/FVC% was recorded in 12,491 (86.9%) patients: median (IQR) 5 (3, 7) measurements. Six thousand five hundred and fifty (52.4%) had persistent AFO, 4507 (36.1%) variable AFO and 1434 (11.5%) absent AFO. Being female, never smoking, having higher BMI or more comorbidities significantly predicted absent and variable AFO. Despite absent AFO, 57% received long-acting bronchodilators and 60% inhaled corticosteroids (50% and 49%, respectively, in those without asthma). In all, 13.1% of patients diagnosed with COPD had unrecorded FEV1/FVC%; 11.5% had absent AFO on repeated measurements, yet many received inhalers likely to be ineffective. Such prescribing is not evidence based and the true cause of symptoms may have been missed.

Scientific fields such as insider-threat detection and highway-safety planning often lack sufficient amounts of time-series data to estimate statistical models for the purpose of scientific discovery. Moreover, the available limited data are quite noisy. This presents a major challenge when estimating time-series models that are robust to overfitting and have well-calibrated uncertainty estimates. Most of the current literature in these fields involve visualizing the time-series for noticeable structure and hard coding them into pre-specified parametric functions. This approach is associated with two limitations. First, given that such trends may not be easily noticeable in small data, it is difficult to explicitly incorporate expressive structure into the models during formulation. Second, it is difficult to know $\textit{a priori}$ the most appropriate functional form to use. To address these limitations, a nonparametric Bayesian approach was proposed to implicitly capture hidden structure from time series having limited data. The proposed model, a Gaussian process with a spectral mixture kernel, precludes the need to pre-specify a functional form and hard code trends, is robust to overfitting and has well-calibrated uncertainty estimates.