Tuberculosis (TB) is an infectious disease which often affects the most vulnerable people in society. The number of new cases per year has been decreasing since 2011. However, there has been no improvement in how long people with TB wait before receiving treatment. The average wait for those with TB in the lungs is 79 days. Those with TB outside the lungs wait even longer. Delays may cause severe illness, spread of TB and even death.
We will carry out four work packages.
One: We will examine electronic health records from CPRD Gold and Aurum and will describe the distribution of TB, the signs and symptoms in those attending GP surgeries and the consequences of illness.
Two: We will compare the distribution of TB in CPRD Gold and Aurum to national TB reports. This will help us understand how similar these two sources of information are.
Three: We will explore the relationship between the timing of TB signs and symptoms and the beginning of treatment. This will help us understand when TB symptoms begin to indicate the presence of disease.
Four: We will develop and test a tool to predict how likely TB is in a person presenting with suggestive symptoms. This tool will be designed to be used in GP surgeries and will identify those who will benefit from referral for investigation.
Overall, the findings of this work will improve patient care by informing Public Health and Primary Care policies, and by providing more details for educational material.
The aim of this piece of work is to explore treatment delays in tuberculosis (TB).
This retrospective cohort study will include adults with active TB diagnosed from July 2011 until the latest available download. There will be nested self-controlled case series studies and a nested case-control study.
Cases will be identified in primary care records, HES and ONS to ensure thorough ascertainment. Symptoms, signs and tests of interest have been identified through a rapid review of the literature and through discussion with a study advisory group. Treatment delay is defined the time from symptom onset to the first appearance of a diagnostic code.
First, we will describe the epidemiology of TB as captured in CPRD Gold and Aurum in terms of demography, site of disease, treatment delay, geography, comorbidities, signs, symptoms, tests and disease outcomes (disease severity and death).
Validation will be carried out by comparing the demography and distribution of cases captured in CPRD Gold and Aurum with details reported in annual TB reports. The differences will be formally tested using t-tests for continuous and chi-squared tests for categorical data.
Self-controlled case series studies will be used to explore the temporal relationships between signs and symptoms and diagnosis. Within-person likelihood ratios for developing a symptom or sign in specified time periods prior to diagnosis will be calculated. A look-back period of five years will ensure that even long treatment delays are captured.
A case-control control study will be used to develop diagnostic prediction models for pulmonary and non-pulmonary TB to inform onward referral from primary care. Logistic regression will be used. Backward selection will be used to select variables within the model. Measures of goodness of fit, calibration, discrimination and clinical utility will be reported. The model will be developed in CPRD Aurum and externally validated in CPRD Gold.
Work package one and two are descriptive and do not have specific outcomes. Variables of interest include: site of disease, age at diagnosis, sex, deprivation, ethnicity, main spoken language, smoking status, evidence of alcohol or drug misuse, evidence of prison stays, evidence of homelessness, being born in a high-risk country and co-morbidities (asthma, COPD, diabetes, renal disease, cancers, silicosis, immunosuppression, being a transplant recipient, gastrectomy and cystic fibrosis). With regards to TB disease, we will also describe the distribution of death due to TB, inpatient admission with TB as the primary reason for admission and markers of illness severity (emergency rather than planned admissions and ICU admissions due to TB).
For the self-controlled case series studies (work package three), the outcome is the relative risk of a symptom or sign in a specified time period in relation to time of diagnosis with TB. Variables related to signs and symptoms will include general (eg fever, malaise, weight loss, fatigue, night sweats) and localised features (eg cough, chest pain, breathlessness, enlarged lymph nodes, joint pain, back pain). Variables related to tests include only those which are ordinarily available in UK primary care (eg full blood count, electrolytes, urine dips, chest x rays).
For the case-control study (work package four), the outcome is a linear predictor to be used to determine the need for onward referral. Variables determining the presence of disease, demography, risk factors, signs, symptoms and tests will all be used.
We acknowledge that some variables will be poorly coded (for example, main spoken language, evidence of prison stays and evidence of homelessness). However, these factors are known to be important in the trajectory of TB disease. As a result, we have erred on the side of sensitivity rather than specificity in the creation of codelists. We will include a missing data category in the descriptions and will use measures reported in annual TB reports to inform implications of missing data.
Krishnarajah Nirantharakumar - Chief Investigator - University of Birmingham
Farah Kidy - Corresponding Applicant - University of Birmingham
Kate Seers - Collaborator - University of Warwick
Krishna Gokhale - Collaborator - University of Birmingham
Nick Parsons - Collaborator - University of Warwick
Shamil Haroon - Collaborator - University of Birmingham
HES Admitted Patient Care;HES Outpatient;ONS Death Registration Data;Practice Level Index of Multiple Deprivation;Patient Level Townsend Index