Teaching (MSc in Public Health @ KCL) & Research Fellow (UCL, KCL & Karolinska Institutet).
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
I'm an epidemiologist with a keen interest in chronic disease epidemiology, health inequalities & methodology. I'm interested in investigating social, ethnic & sexual determinants of health & their development across the life-course. Main chronic diseases include CVD, diabetes & their risk factors, & their development in childhood, adolescence and early adulthood, mental health & multimorbidity). Other interests include perinatal and pregnancy related outcomes and cancer epidemiology.
March 2019 - March 2021
- Research Associate
- Lead researcher on projects examining lifecourse development of comorbidity and multimorbidity in the UK birth cohorts (NSHD46, NCDS58, BCS70, MCS). Principle investigator on a mixed methods study examining health (mental health, general health & health-related behaviours) in sexual and ethnic minority adolescents & young people.
January 2010 - March 2014
- Parental influences on CVD risk factors in children
- CVD, SEP, Overweight & obesity, children, Cholesterol, LDL/HDL, Lifestyle, Smoking, Physical activity, Alcohol consumption
Background: Ethnic minorities/immigrants have differential health as compared with natives. The epidemic in child overweight/obesity (OW/OB) in Sweden is leveling off, but lower socioeconomic groups and immigrants/ethnic minorities may not have benefited equally from this trend. We investigated whether nonethnic Swedish children are at increased r...
Ethnic minorities/immigrant groups tend to have increased risk for preterm birth. Less is known about this risk in diverse immigrant groups, couples of mixed ethnic-origin and in relation to duration of residence. Data from the Swedish Medical Birth Register on 1,028,303 mothers who gave birth to 1,766,026 singleton live born infants (1982-2002), w...
Background The impact of ethnicity and socio-economic status (SES) on glycaemic control during childhood Type 1 diabetes is poorly understood in England and Wales. Methods We studied 18 478 children with Type 1 diabetes (< 19 years) attending diabetes clinics and included in the 2012–2013 National Paediatric Diabetes Audit. Self-identified ethnicit...
Background The aim was to investigate associations between different measures of socioeconomic position (SEP) and incidence of brain tumours (glioma, meningioma and acoustic neuroma) in a nationwide population-based cohort. Methods We included 4 305 265 individuals born in Sweden during 1911–1961, and residing in Sweden in 1991. Cohort members were...
Background We aimed to estimate multimorbidity trajectories and quantify socioeconomic inequalities based on childhood and adulthood socioeconomic position (SEP) in the risks and rates of multimorbidity accumulation across adulthood. Methods and findings Participants from the UK 1946 National Survey of Health and Development (NSHD) birth cohort st...
Introduction According to the UN Refugee Agency (UNHCR), around 40% of the 79.5 million forcibly displaced persons in the end of the year 2019 were children. Exposure to violence and mental health problems such as posttraumatic stress disorder are frequently reported among migrant children, but there is a knowledge gap in our understanding of the c...
Aims To examine inequalities related to dual sexual- and ethnic-identities in risk for health, wellbeing, and health-related behaviours in a nationally representative sample of adolescents. Methods 9,789 adolescents (51% female) aged 17 years from the UK-wide Millennium Cohort Study, with data on self-identified sexual- and ethnic-identities. Adole...
Aim To examine socioeconomic inequalities in comorbidity risk for overweight (including obesity) and mental ill-health in two national cohorts. We investigated independent effects of childhood and adulthood socioeconomic disadvantage on comorbidity from childhood to mid-adulthood, and differences by sex and cohort. Methods Data were from 1958 Nati...
Aim To examine socioeconomic inequalities in risk of comorbidity between overweight (including obesity) and mental ill-health in two national cohorts. We investigated independent effects of childhood and adulthood socioeconomic disadvantage on comorbidity from childhood to mid-adulthood, and differences by sex and cohort. Methods Data were from 1...
This study examined socioeconomic inequalities in multimorbidity across adulthood and into old age using data from the MRC 1946 National Survey of Health & Development (NSHD). The study was restricted to participants that attended the age 36 assessment in 1982 and any one of the follow-up assessments at ages 43, 53, 63 and 69. We analysed 1. asso...
Objective: To estimate multimorbidity trajectories and quantify socioeconomic inequalities based on both childhood and adulthood socioeconomic position in the risks and rates of adult multimorbidity accumulation. Design: Prospective longitudinal national birth cohort study. Methods: Participants from the 1946 National Survey of Health and Developme...
Background Some studies hypothesize that birth month—as a proxy of exposure to ultraviolet radiation in early infancy—is associated with increased risk of skin tumors. Methods We studied a national cohort of all 5 874 607 individuals born in Sweden to parents of Swedish or Nordic origin as a proxy for Caucasian origin, 1950 to 2014. The cohort was...
Background Multimorbidity (≥2 chronic diseases) is increasingly prevalent in ageing populations and presents a public health challenge in successful disease management. Most evidence for multimorbidity at different ages comes from cross-sectional data, hindering understanding the extent and types of multimorbidity across the lifecourse, how they de...
Background For childhood onset type 1 diabetes (T1D), the pathogenesis of atherosclerosis is greatly accelerated and results in early cardiovascular disease (CVD) and increased mortality. However, cardioprotective interventions in this age group are not routinely undertaken. Aims To document prevalence of cardiovascular risk factors from diagnosis...
Background: Not much is known about glycaemic-control trajectories in childhood-onset type 2 diabetes (T2D). We investigated characteristics of children and young people (CYP) with T2D and inequalities in glycemic control. Methods: We studied 747 CYP with T2D, <19 years of age in 2009-2016 (from the total population-based National Pediatric Diab...
This animation shows - through the eyes of children - the results from one of our publications which reported the steep increased prevalence of type 2 diabetes in children & adolescents belonging to ethnic minority backgrounds in England & Wales. The paper can be accessed here: https://www.ncbi.nlm.nih.gov/pubmed/27426206
It is unclear whether diabetic ketoacidosis (DKA) severity at diagnosis affects the natural history of type 1 diabetes (T1D). We analysed associations between DKA severity at diagnosis and glycaemic control during the first year post-diagnosis. We followed 341 children with T1D, <19 years (64% non-white) attending paediatric diabetes clinics in Eas...
Objective: To determine whether the Institute Of Medicine's (IOM) 2009 guidelines for weight-gain during pregnancy are predictive of maternal and infant outcomes in ethnic minority populations. Methods: We designed a population-based study using administrative data on 181,948 women who delivered live singleton births in Washington State between 200...
Background Ethnic minorities with Type 1 Diabetes (T1D) have worse glycaemic control (higher glycated haemoglobin or HbA1c) and greater risk for vascular complications. However, the impact of ethnicity on early glycaemic control when patients experience transient remission post-diagnosis is limited. We used modelling techniques to examine the indep...
Background Socioeconomic disparities in mortality from various cancers are well documented with patients from lower socioeconomic background having an increased risk for mortality. However, similar evidence on differences in mortality from tumours of the central nervous system is both limited and conflicting. We investigated associations between so...
Some ethnic minorities with type 1 diabetes (T1D) have worse glycemic control (higher glycated hemoglobin (HbA1c)) and increased risk for vascular complications. There is limited evidence on the impact of ethnicity on early glycemic control when most patients experience transient remission postdiagnosis. We examined associations between ethnicity a...
BACKGROUND: The impact of ethnicity and socio-economic status (SES) on glycaemic control during childhood Type 1 diabetes is poorly understood in England and Wales. METHODS: We studied 18 478 children with Type 1 diabetes (< 19 years) attending diabetes clinics and included in the 2012–2013 National Paediatric Diabetes Audit. Self-identified ethnic...
Background: There is marked variation in diabetes outcomes for children and adolescents across the UK. We used modelling techniques to examine the independent contributions of deprivation, ethnicity, insulin pump use, and health service use on HbA1c trajectories across adolescence. Methods: Prospective data from a large UK Paediatric & Adolescen...
Purpose: Ethnic minority children are at a greater risk for type 2 diabetes (T2D). However, current prevalence of T2D among children and young people is unknown in England and Wales. In addition, little is known on glycemic control in pediatric T2D globally. Methods: Using data from the National Paediatric Diabetes Audit for 2012-2013 with >98%...
Background Ethnic minority children are at greater risk for type 2 diabetes. The current prevalence of type 2 diabetes in children in England and Wales is not known. Additionally, very little is known on glycaemic control in paediatric type 2 diabetes globally. Methods Using data from the National Paediatric Diabetes Audit (NPDA) for 2012–13, we es...
Background: Landmark studies in adult-onset type 1 diabetes (T1D) populations indicate that improved glycaemic control through use of intensive insulin therapy is strongly associated with reduced risk for the development of diabetes-related complications and mortality in later years. However, it is unclear whether these associations can be extrapo...
Background: The impact of ethnicity and socio-economic status (SES) on glycaemic control during childhood Type 1 diabetes is poorly understood in England and Wales. Methods: We studied 18 478 children with Type 1 diabetes (< 19 years) attending diabetes clinics and included in the 2012-2013 National Paediatric Diabetes Audit. Self-identified eth...
Health inequalities in paediatric type 1 diabetes care in England and Wales. Assessing the impact of ethnicity and socioeconomic status.
Aims We investigated whether the UK has higher child and youth diabetes mortality than in comparable European countries and the USA. Methods We obtained data from the WHO World Mortality Database for the UK, the USA and the EU15+ (the 15 countries of the EU in 2004 plus Australia, Canada and Norway) for 1990 to 2010. Diabetes mortality rates were...
Background: Precursors of cardiovascular diseases (CVD) originate in childhood. We investigated relationships of children's CVD risk factors with parent's socio-economic position (SEP) and lifestyle and how CVD risk factors correlate within families. Methods: We studied 602 families with 2141 individuals comprising two full sibs; aged 5-14 years...
To study the relationship of children’s cardiovascular (CVD) risk factors with parent’s CVD risk factors (RF), socioeconomic position (SEP) and lifestyle. 602 families (2141 individuals), with two full sibs; aged 5-14 years and their biological parents (Uppsala Family Study) formed the study population. Cholesterol, ApoB/ApoA1 ratio, adiponectin, b...
Aim: To investigate the role of adiponectin in the association between birth weight (BW) and insulin resistance in elderly men. Methods: We studied 727 men born 1920-24 and resident in Uppsala, Sweden, in 1970, who were part of the ULSAM cohort study with more than 38 years follow-up. Information on serum adiponectin, clinical measures of insulin r...
Background Social status is associated with cardiovascular disease (CVD) prevalence and incidence. We aimed to study relationships between i) socioeconomic position (SEP) and common CVD biomarkers; cholesterol, LDL/HDL, ApoB/ApoA1 and adiponectin ii) SEP and CVD mortality in a Swedish-population-based sample, and to assess if these associations cha...
Background Precursors of cardiovascular diseases (CVD) originate in childhood. We investigated the relationship of children's CVD risk-factors with parent's socioeconomic position (SEP) and lifestyle. We also studied how CVD risk-factors correlate within families. Methods We studied 602 families (2141 individuals) comprising two full sibs; aged 5–...
Social status is associated with cardiovascular disease (CVD) prevalence and incidence. Aims: to investigate relationships between socioeconomic position (SEP) and common CVD biomarkers including adiponectin not previously investigated in a Swedish-population sample, and to assess if these associations changed with age. Population-based longitudina...
My dataset is longitudinal in the long format, and each individual has 5 rows of data with each row representing one wave (i.e. total 5 waves).
I just want to create a variable called year which can take on the same 5 values: 1982, 1989, 1999, 2009 & 2015 for subjects and is understood by Stata to be in calendar years. I will use this year variable to help create other variables like duration and age etc.
I would like the year variable to look this:
id year 1 1982 1 1989 1 1999 1 2009 1 2015 2 1982 2 1989 2 1999 2 2009 2 2015 3 1982 3 1989 3 1999 3 2009 3 2015
Any ideas how to generate the above year variable in Stata date language?
I'm about to impute missing data in a longitudinal dataset that follows a cohort of subjects over 50 years with data collected at 5 time points (5 waves).
I use Stata for data analysis and I've previously analysed longitudinal data in the 'long format'. This is relatively easy & straight forward in Stata.
However, it is recommended to impute longitudinal data in the 'wide format' and then reshape the data back to long format for analysis. I find this rather cumbersome, as it requires a fair bit of data prep. Imputing in Stata add 'prefixes' to all imputed variables (for example) which can make re-shaping the data difficult.
1. Is it possible to impute longitudinal data in the long format? pros & cons?
2. Is there an alternative or easier strategy when imputing longitudinal data using Stata?
Thoughts, ideas, recommendations welcome!
I'm analyzing the association between y (a clinical variable) and time (time since diagnosis of diabetes), adjusting for gender and at age at diagnosis. We already know from existing literature and the raw data, that y has a non-linear relationship with time since diagnosis. I've fit a multilevel piece-wise linear spline model to account for the non-linearity of the data with knots at 1, 2, 4, 8, 12, 17, 24 & 34 months after diagnosis (total follow-up time 72 months). The data is longitudinal with subjects having multiple measurements of Y, so essentially this is a growth curve analysis (random slope & random coefficient model):
The Stata code I used:
. mkspline durm1 1 durm2 2 durm4 4 durm8 8 durm12 12 durm17 17 durm24 24 durm34 34 durm36 = durationm
. mixed hba1cmol durm1 durm2 durm4 durm8 durm12 durm17 durm24 durm34 durm36 sexnew diagagenew || id: durationm, cov(unstr) mle
I would now like to visualize the data post regression - plot y with time since diagnosis first for all subjects and then separately for males and females. What is the best way to do this?
I've been told margins and marginsplot (which I generally use) doesn't work too well for linear spline models. Is this true? There is a user written package to help plot data post regression when using restricted cubic splines (the 'postrcspline' package).
Is there a way to plot what I would like to see? Due to the nature of the data, it is stored on a secure online server and I'm unable to download and run any user-written commands which complicates matters.....
I'm running a set of mixed effects models to analyse change in my outcome after diagnosis with a disease, given a set of predictors. We know that the outcome does not have a linear relationship with time (duration since diagnosis). Models thus require quadratic and cubic terms in addition to the linear term for duration.
I know the number of polynomials should be one less than the average number of measurements of the outcome per subject (N-1).
In my case, I have on average 3.5 outcome measurements/subject and three terms for duration - hence I'm 'breaking the rule'.
So, how strict is this rule? And how does it affect model fit if one does not stick to the N-1 recommendation? My models seems to run well with no complaints though.
I'm running a series of multilevel regression models (mixed effects or random coefficient analysis) in Stata 13 to investigate associations between a set of predictors, time (here interpreted as duration in months from time of diagnosis) and my outcome of interest which is continuous (say cholesterol in mmol/L).
The main purpose is to investigate rate of change (i.e. this is a longitudinal analysis) - does cholesterol change with duration (time) given a set of certain predictors.
I know that the modeling results in two parts; the fixed effects part and the random effects part. I know how to interpret the fixed effects part, but could someone help me understand the estimates from the random effects part, when this is run for longitudinal analysis?
Below, we see that cholesterol reduces by -19 units per month (duration is in months). Mixed and black ethnicities have cholesterol levels at 3.4 & 3.3 at time 0 (i.e. at diagnosis here) and so on. But how do I interpret the estimates under 'random-effects parameters', in the bottom half of the output?
Example of output:
xtmixed hba1cifcc2 durationm durationm2 sex1 i.ethnicnew2 || id: durationm, cov(unstr) mle var, if diagyr>2004 & durationm<6.1 & imd4!=.
Mixed-effects ML regression Number of obs = 1028
Group variable: id Number of groups = 443
Obs per group: min = 1
avg = 2.3
max = 7
Wald chi2(9) = 729.89
Log likelihood = -4329.3449 Prob > chi2 = 0.0000
cholesterol | Coef. Std. Err. z P>|z| [95% Conf. Interval]
durationm | -19.43518 .7736272 -25.12 0.000 -20.95146 -17.9189
durationm2 | 2.402219 .1226445 19.59 0.000 2.16184 2.642598
sex| -1.840137 1.352356 -1.36 0.174 -4.490706 .8104314
mixed | 3.442956 2.547798 1.35 0.177 -1.550636 8.436549
Black | 3.286653 2.12706 1.55 0.122 -.8823077 7.455614
asian | 5.825651 1.820642 3.20 0.001 2.257258 9.394044
_cons | 93.19885 2.935992 31.74 0.000 87.44441 98.95329
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
id: Unstructured |
var(durati~m) | 21.35454 2.9492331 16.03032 27.2855
var(_cons) | 352.785 38.02899 284.87269 434.91405
corr(durati~m,_cons) | -69.80694 9.286172 -88.562195 -50.74312
sd(Residual) | 10.55247 .4243647 9.752664 11.41787
LR test vs. linear regression: chi2(3) = 214.13 Prob > chi2 = 0.0000
I'm running a set of multilevel linear regression models to analyze associations between x and y using longitudinal data taking into account a set of predictors. My hypothesis is that x changes with y (here y is duration and my time variable).
I do know that the relationship between x and y is not completely linear. So, I introduced a quadratic term for duration to account for the non-linearity. However, on average, each subjects only has 2 observations - and I assume that a mean of 2 observations per subject could be problematic when having a quadratic term as this would require at least 3-4 observations per subject on average?
If the above is true, then is it ok to run the model with just linear duration despite knowing that the relationship between x and y (duration) isn't completely linear?
Does anyone have the Stata code to calculate overweight and obesity in children (<18 years) according to WHOs 2007 growth curve cut-offs (85th and 95th percentiles)?
Would be really great (and time saving!) if anyone wouldn't mind sharing their code for the above.
I'm currently analyzing ethnic differences in an outcome using data on several thousands of children that have between 1 and 5 clinic visits (mean 2.9 clinic visits) during a calender year. These children are 'clustered' within 178 clinics. However, missing data is an issue. I thought of using multiple imputation to overcome the problem of missing data. However, if I'm right imputation isn't recommended when data is clustered? Is this right?
I would like to know if it's appropriate to plot a histogram based on estimates (RRR) obtained from a multinomial logistic regression model?
From what I understand the the estimates for all categories of the outcome variable in a multinomial logistic regression model should add up to one and I was told that a histogram is probably not the best choice in such a case. Would a forest plot be a better option?
I'm trying to construct a forest plot using RRR from a multinomial logistic regression model in Stata. Does anyone have any tips on how this can be done using the metan command in Stata (or any other command)? While the metan command is constructed to plot the estimates from a meta-analysis, I'm assuming this can also be used to plot estimates from a regression model.
Substantial evidence shows that young-people and -adults who identify as sexual and gender minorities (Lesbian, Gay, Bisexual, Transgender or Queer [LGBTQ+]) have substantially poorer mental health (depression, anxiety and suicidality) and well-being, and are at increased risk of self-harm including substance abuse and risky behaviour compared to cisgender heterosexual peers. They are more likely to be bullied or harassed, face stigma and forced to hide their sexual and gender identities. This is despite the considerable sociocultural shifts in better understanding sexual and gender identities coupled with greater political rights and significant change in societal attitudes and acceptance of sexual minorities in recent decades in many parts of the world. To date most studies on mental health in LGBTQ+ individuals were largely conducted on White individuals with the few studies on LGBTQ+ and ethnic minorities originating in the US (based on smaller studies, some with discordant results). However, in the UK there is little knowledge on mental health in ethnic minority individuals who also identify as LGBTQ+ (i.e. multiple minority identities) and in general how intersecting oppressions and privileges related to sexual- and gender-identities and ethnicity impact health. This critical gap in knowledge is remarkable as there is substantial evidence of poorer health outcomes in many ethnic minorities. Ethnic minorities are more like to face bullying, harassment and racism as a result of their ethnic-origin or religion, regardless of their sexual or gender identity. Thus ethnic minorities could be at an even higher risk for poorer mental health due to the cumulative effects of two or more minority identities. Poor mental and physical health especially chronic conditions in adolescence or early adulthood can have long-term consequences as they can impact access to higher education, the labour market, effective and positive participation in society, and track across adulthood increasing risk for morbidity. This project will examine the intersection of multiple minority identities (sexual-gender-ethnic-religious) to better understand diversity in the lived experiences of LGBTQ+ groups and their risk in regard to mental health and wellbeing. This project is based on existing national quantitative data and qualitative data that will be collected over period of 4 months.
Are ethnic minority children and young adults (CYP) and those from lower socioeconomic backgrounds at greater risk for type 2 diabetes? Are they also at increased risk for poorer diabetes management (worse glycaemic control)? How does the clinical profile of ethnic minority children with type 2 diabetes differ from that in White children? We plan to investigate the above research questions using a large nationally representative cohort of children and young people diagnosed with type 2 diabetes over the past ten years in England and Wales.