Search

IntroductionThis study considers the prediction of the time until two survival outcomes have both occurred. We compared a variety of analytical methods motivated by a typical clinical problem of multimorbidity prognosis.MethodsWe considered five methods: product (multiply marginal risks), dual-outcome (directly model the time until both events occur), multistate models (msm), and a range of copula and frailty models. We assessed calibration and discrimination under a variety of simulated data scenarios, varying outcome prevalence, and the amount of residual correlation. The simulation focused on model misspecification and statistical power. Using data from the Clinical Practice Research Datalink, we compared model performance when predicting the risk of cardiovascular disease and type 2 diabetes both occurring.ResultsDiscrimination was similar for all methods. The product method was poorly calibrated in the presence of residual correlation. The msm and dual-outcome models were the most robust to model misspecification but suffered a drop in performance at small sample sizes due to overfitting, which the copula and frailty model were less susceptible to. The copula and frailty model's performance were highly dependent on the underlying data structure. In the clinical example, the product method was poorly calibrated when adjusting for 8 major cardiovascular risk factors.DiscussionWe recommend the dual-outcome method for predicting the risk of two survival outcomes both occurring. It was the most robust to model misspecification, although was also the most prone to overfitting. The clinical example motivates the use of the methods considered in this study.

Introduction: A patient is eligible for statins in England if they have a 10-year risk of cardiovascular disease >10%. We hypothesize that if statin discontinuation rates are high it may be better to delay statin initiation until patients are at a higher risk, to maximize the benefit of the drug. Methods: A four-state health state transition model was used to assess the optimal time to initiate statins after a risk assessment, in order to prevent the highest number of cardiovascular events, for a given risk profile (age, gender, risk) and adherence rate. A Clinical Practice Research Datalink dataset linked to Hospital Episodes Statistics and Office for National Statistics was used to inform the transition probabilities in this model, taking into account observed statin discontinuation and re-continuation patterns. Results: Our results suggest, if statins are initiated in a cohort of 50-year old men with a 10% 10-year risk, we prevent 4.78 events per 100 individuals. If we wait 10 years to prescribe, at which point 10-year risk scores are at 20%, we prevent 5.45 events per 100 individuals. If the observed discontinuation rate was reduced by a sixth, third or half in the same cohort, we would prevent 7.29, 9.01 or 10.22 events per 100 individuals. Conclusions: In certain scenarios, extra cardiovascular disease events could be prevented by delaying statin initiation beyond a risk of 10% until reaching a age (59 for men, 63 for women), based on statin discontinuation rates in England. The optimal time to initiate statins was driven by age, not by cardiovascular risk.

© 2020 The Authors. Learning Health Systems published by Wiley Periodicals, Inc. on behalf of the University of Michigan Background: The “learning health system” has been proposed to deliver better outcomes for patients and communities by analyzing routinely captured health information and feeding back results to clinical staff. This approach has been piloted in the Connected Health Cities (CHC) programme in four regions in the North of England. This paper presents the results of the evaluation of this program conducted between February and December 2018. Methods: Fifty nine semistructured interviews were completed with a mix of CHC programme staff and external partners who had contributed to the delivery of the CHC programme. Interviews were audio recorded and transcribed verbatim. This also included the review of project documentation including project reports and minutes of project group meetings, in addition to a short online survey that was completed by 31 members of CHC programme staff. Data were analyzed thematically. Results: Two overarching themes emerged through the thematic analysis of participant interview: (a) challenges in the implementation of learning health system pathways, and (b) benefits to the CHC approach for both staff and patients. In particular, time constraints in delivering an ambitious program of work, data quality, and accessibility, as well as the long-term sustainability of the CHC programme were noted as key challenges in implementing a LHS at scale. Conclusions: The findings from this evaluation provide valuable insight into creating learning health system at scale, including the potential benefits and likely challenges.

Purpose: To assess the impact of a variety of methodological parameters on the association between six drug classes and five key adverse events in multiple databases. Methods: The selection of Drug-Adverse Event pairs was based on public health impact, regulatory relevance, and the possibility to study a broad range of methodological issues. Common protocols and data analytical specifications were jointly developed and independently and blindly executed in different databases in Europe with replications in the same and different databases. Results: The association between antibiotics and acute liver injury, benzodiazepines and hip fracture, antidepressants and hip fracture, inhaled long-acting beta2-agonists and acute myocardial infarction was consistent in direction across multiple designs, databases and methods to control for confounding. Some variation in magnitude of the associations was observed depending on design, exposure and outcome definitions, but none of the differences were statistically significant. The association between anti-epileptics and suicidality was inconsistent across the UK CPRD, Danish National registries and the French PGRx system. Calcium channel blockers were not associated with the risk of cancer in the UK CPRD, and this was consistent across different classes of calcium channel blockers, cumulative durations of use up to >10years and different types of cancer. Conclusions: A network for observational drug effect studies allowing the execution of common protocols in multiple databases was created. Increased consistency of findings across multiple designs and databases in different countries will increase confidence in findings from observational drug research and benefit/risk assessment of medicines.

PURPOSE: Results from observational studies on the same exposure-outcome association may be inconsistent because of variations in methodological factors, clinical factors or health care systems. We evaluated the consistency of results assessing the association between antidepressant use and the risk of hip/femur fractures in three European primary care databases using two different study designs. METHODS: Cohort and nested case control studies were conducted in three European primary care databases (Spanish BIFAP, Dutch Mondriaan and UK THIN) to assess the association between use of antidepressants and hip/femur fracture. A common protocol and statistical analysis plan was applied to harmonize study design and conduct between data sources. RESULTS: Current use of antidepressants was consistently associated with a 1.5 to 2.5-fold increased risk of hip/femur fractures in all data sources with both designs, with estimates for SSRIs generally higher than those for TCAs. In general, risk estimates in Mondriaan, the smallest data source, were higher compared to the other data sources. This difference may be partially explained by an interaction between SSRI and age in Mondriaan. Adjustment for GP-recorded lifestyle factors and matching on general practice had negligible impact on adjusted relative risk estimates. CONCLUSION: We found a consistent increased risk of hip/femur fracture with current use of antidepressants across different databases and different designs. Applying similar pharmacoepidemiological study methods resulted in similar risks for TCA use and some variation for SSRI use. Some of these differences may express real (or natural) variance in the exposure-outcome co-occurrences. Copyright © 2016 John Wiley & Sons, Ltd.

The Pharmacoepidemiological Research on Outcomes of Therapeutics by a European ConsorTium (PROTECT) initiative was a collaborative European project that sought to address limitations of current methods in the field of pharmacoepidemiology and pharmacovigilance. Initiated in 2009 and ending in 2015, PROTECT was part of the Innovative Medicines Initiative, a joint undertaking by the European Union and pharmaceutical industry. Thirty-five partners including academics, regulators, small and medium enterprises, and European Federation of Pharmaceuticals Industries and Associations companies contributed to PROTECT. Two work packages within PROTECT implemented research examining the extent to which differences in the study design, methodology, and choice of data source can contribute to producing discrepant results from observational studies on drug safety. To evaluate the effect of these differences, the project applied different designs and analytic methodology for six drug-adverse event pairs across several electronic healthcare databases and registries. This papers introduces the organizational structure and procedures of PROTECT, including how drug-adverse event and data sources were selected, study design and analyses documents were developed, and results managed centrally.

PURPOSE: UK primary care databases are frequently used in observational studies with cancer outcomes. We aimed to systematically review methods used by such studies to identify and validate incident cancers of the breast, colorectum, and prostate. METHODS: Medline and Embase (1980-2013) were searched for UK primary care database studies with incident breast, colorectal, or prostate cancer outcomes. Data on the methods used for case ascertainment were extracted and summarised. Questionnaires were sent to corresponding authors to obtain details about case ascertainment. RESULTS: Eighty-four studies of breast (n = 51), colorectal (n = 54), and prostate cancer (n = 31) were identified; 30 examined >1 cancer type. Among the 84 studies, 57 defined cancers using only diagnosis codes, while 27 required further evidence such as chemotherapy. Few studies described methods used to create cancer code lists (n = 5); or made lists available directly (n = 5). Twenty-eight code lists were received on request from study authors. All included malignant neoplasm diagnosis codes, but there was considerable variation in the specific codes included which was not explained by coding dictionary changes. Code lists also varied in terms of other types of codes included, such as in-situ, cancer morphology, history of cancer, and secondary/suspected/borderline cancer codes. CONCLUSIONS: In UK primary care database studies, methods for identifying breast, colorectal, and prostate cancers were often unclear. Code lists were often unavailable, and where provided, we observed variation in the individual codes and types of codes included. Clearer reporting of methods and publication of code lists would improve transparency and reproducibility of studies. Copyright © 2014 John Wiley & Sons, Ltd.

PURPOSE: Prior event rate ratio (PERR) adjustment method has been proposed to control for unmeasured confounding. We aimed to assess the performance of the PERR method in realistic pharmacoepidemiological settings. METHODS: Simulation studies were performed with varying effects of prior events on the probability of subsequent exposure and post-events, incidence rates, effects of confounders, and rate of mortality/dropout. Exposure effects were estimated using conventional rate ratio (RR) and PERR adjustment method (i.e. ratio of RR post-exposure initiation and RR prior to initiation of exposure). RESULTS: In the presence of unmeasured confounding, both conventional and the PERR method may yield biased estimates, but PERR estimates appear generally less biased estimates than the conventional method. However, when prior events strongly influence the probability of subsequent exposure, the exposure effect from the PERR method was more biased than the conventional method. For instance, when the effect of prior events on the exposure was RR = 1.60, the effect estimate from the PERR method was RR = 1.13 and from the conventional method was RR = 2.48 (true exposure effect, RR = 2). In all settings, the variation of the estimates was larger for the PERR method than for the conventional method. CONCLUSION: The PERR adjustment method can be applied to reduce bias as a result of unmeasured confounding. However, only in particular situations, it can completely remove the bias as a result of unmeasured confounding. When applying this method, theoretical justification using available clinical knowledge for assumptions of the PERR method should be provided. Copyright © 2014 John Wiley & Sons, Ltd.

AIM: The aim was to evaluate clinical risk factors associated with myotoxicity in statin users. METHODS: This was a cohort study of patients prescribed a statin in UK primary care practices contributing to the Clinical Practice Research Datalink. Outcomes of interest were creatine phosphokinase (CPK) concentrations and clinical records of rhabdomyolysis. RESULTS: The cohort comprised 641,703 statin users. Simvastatin was most frequently prescribed (66.3%), followed by atorvastatin (24.4%). CPK was measured in 127,209 patients: 81.4% within normal range and 0.7% above four times ULN CPK compared with normal CPK (OR 1.28, 95% CI 1.01, 1.60). Rosuvastatin users had higher risk of >four times ULN CPK (OR 1.62, 95% CI 1.22, 2.15) as did patients with larger daily doses of other statin types. A recent clinical record of myalgia was associated with an increased OR of >four times ULN CPK (OR 1.73, 95% CI 1.37, 2.18). In patients who were rechallenged to statins and had repeat CPK measurements after >four times ULN CPK abnormalities, 54.8% of the repeat CPK values were within normal range, 32.1% between one to three times and 13.0% >four times ULN. CONCLUSIONS: The frequencies of substantive CPK increases and rhabdomyolysis during statin treatment were low, with highest risks seen in those on large daily doses or interacting drugs and on rosuvastatin. CPK measurements appeared to have been done in a haphazard manner and better guidance is needed.

PURPOSE: Difficulties may be encountered when undertaking a benefit-risk assessment for an older product with well-established use but with a benefit-risk balance that may have changed over time. This case study investigates this specific situation by applying a formal benefit-risk framework to assess the benefit-risk balance of warfarin for primary prevention of patients with atrial fibrillation. METHODS: We used the qualitative framework BRAT as the starting point of the benefit-risk analysis, bringing together the relevant available evidence. We explored the use of a quantitative method (stochastic multi-criteria acceptability analysis) to demonstrate how uncertainties and preferences on multiple criteria can be integrated into a single measure to reduce cognitive burden and increase transparency in decision making. RESULTS: Our benefit-risk model found that warfarin is favourable compared with placebo for the primary prevention of stroke in patients with atrial fibrillation. This favourable benefit-risk balance is fairly robust to differences in preferences. The probability of a favourable benefit-risk for warfarin against placebo is high (0.99) in our model despite the high uncertainty of randomised clinical trial data. In this case study, we identified major challenges related to the identification of relevant benefit-risk criteria and taking into account the diversity and quality of evidence available to inform the benefit-risk assessment. CONCLUSION: The main challenges in applying formal methods for medical benefit-risk assessment for a marketed drug are related to outcome definitions and data availability. Data exist from many different sources (both randomised clinical trials and observational studies), and the variability in the studies is large. Copyright � 2014 John Wiley & Sons, Ltd.

AIMS: Electronic healthcare records (EHRs) are increasingly used to store clinical information. A secondary benefit of EHRs is their use, in an anonymized form, for observational research. The Clinical Practice Research Datalink (CPRD) contains EHRs from primary care in the UK and, despite 1083 peer-reviewed research publications, has never been used to obtain pharmacogenetic samples. Using a statin-induced myopathy paradigm, we evaluated using the CPRD to obtain patient samples for a pharmacogenetic study targeting 250 cases and 500 controls from UK general practitioner (GP) practices. METHODS: The CPRD identified potential patients fitting specific case-definition criteria (active rhabdomyolysis or creatine phosphokinase > four times the upper limit of normal), and corresponding GP practices were asked to invite patient participation. Consenting patients were requested to provide either saliva or blood samples and to complete an ethnicity questionnaire. Control subjects were recruited from the same GP practice (saliva) or a small number of practices (blood). Samples were forwarded for DNA extraction. RESULTS: Thirty-six months of recruitment yielded DNA samples from 149 statin-induced myopathy cases and 587 tolerant controls. Data show that contacting patients through their GP is a reliable method for obtaining samples without compromising anonymity. Saliva collection directly from patients was considerably less effective than blood sampling. After 10 months of recruitment, saliva sampling was suspended in favour of blood sampling. CONCLUSIONS: We demonstrate the potential of EHRs for identifying accurately phenotyped cases and controls for pharmacogenetic studies. Recruitment was successful only because of the willingness of GP practices to participate and the existence of strong doctor-patient relationships. The present study provides a model that can be implemented in future genetic analyses using EHRs.

Oral anticoagulation (OAC) is used to prevent stroke in patients with atrial fibrillation or heart valve disorders. However, OAC may cause haemorrhagic stroke. We completed a randomized trial of stroke secondary prevention 1, using primary care electronic health records (EHRs) collected into the UK Clinical Practice Research Datalink (CPRD). A small proportion of participants were prescribed warfarin following a haemorrhagic stroke diagnosis. The present research aimed to explain this observation. Participants were selected from all 11 391 patients with prevalent stroke, from 106 family practices in England and Scotland, included in the trial 1. Patients were included if they had a record of haemorrhagic stroke before the trial and were treated with warfarin during the trial. Family physicians were sent a postal questionnaire asking respondents about the nature of the stroke suffered by the patient and indications for warfarin therapy. Type of stroke was coded using the International Classification of Diseases 10th Revision 2. Warfarin accounts for 99% of OAC prescriptions in the CPRD. The project was approved by an external scientific and ethical review committee (ISAC protocol 08_083A2). GPs consented to their practice's participation. Fully anonymized data were used and no patient consent was sought. There were 134 (1.2%) participants with EHRs documenting previous haemorrhagic stroke who were prescribed warfarin during the trial. The index stroke event was coded into the EHR as a sub-arachnoid haemorrhage for 25 (24%) of cases and as intracerebral haemorrhage (ICH) for 78 (76%) of cases. There were 40 (47%) women and the mean age was 75 years (range 40 to 94 years). Questionnaire responses were obtained for 103 (77%) patients with sufficient information to confirm the type of stroke obtained for 86 (64%) (Figure 1). Figure 1 Flow chart showing physician-reported data for confirmed stroke type and indication for OAC therapy in 103 participants with electronic records for haemorrhagic stroke Questionnaire responses revealed that there were 41 (48%) patients with a physician-confirmed diagnosis of cerebral infarction (n = 39) or other non-haemorrhagic stroke (n = 2). The diagnosis was confirmed by imaging reports recorded for 36 (88%) of patients. There were five patients with a code for ischaemic stroke recorded after the initial haemorrhagic stroke but only two of these were within 1 year of the index stroke, possibly representing a clarification of the initial diagnosis. Indications for OAC in this group of participants included atrial fibrillation (n = 27), disease of pre-cerebral and cerebral arteries (n = 4), venous thromboembolism and other and not known (n = 10). There were 45 patients with a physician-confirmed diagnosis of haemorrhagic stroke including sub-arachnoid haemorrhage (n = 17) and ICH (n = 28). The diagnosis was supported by imaging results for 38 (84%) patients. Indications for OAC therapy included atrial fibrillation (n = 20), venous thrombosis (n = 10), pulmonary embolism (n = 5), aortic valve replacement (n = 3), percutaneous coronary intervention, vasculitis and not known (n = 7). We conclude that in the EHRs of a large population of stroke patients, there may be appreciable numbers with a diagnosis of haemorrhagic stroke who are treated with OAC therapy. We reported previously on the clinical coding of stroke in EHRs 3 and the potential for misclassification of stroke diagnoses. The present results should not be considered to evaluate the overall reliability of stroke coding in EHRs, because the sample was one in which data discrepancies were expected. In about half of patients, the stroke event was mis-coded as a haemorrhagic stroke when the physician-reported diagnosis for the same event date was one of cerebral infarction. Misclassification might sometimes result from haemorrhagic transformation, which is not necessarily a contra-indication to OAC therapy. Physicians may draw on sources of data from outside the EHR. Nevertheless, it is concerning that appreciable numbers of stroke patients are prescribed OAC therapy with incorrect diagnostic information coded into their EHRs, with potential medico-legal implications. Improved information sharing between providers would be of benefit to physicians. A second group comprises patients with a confirmed diagnosis of haemorrhagic stroke who were treated with OAC therapy for co-existing thromboembolic disorders. This highlights the difficult balance of risks and benefits that may sometimes be required in clinical practice. However, the use of OAC therapy in patients with previous ICH remains controversial 4. Improved evidence and decision support are required for these high risk patients.

Purpose Large electronic datasets are increasingly being used to evaluate healthcare delivery. The aim of this study was to compare information held by cancer registries with that of the General Practice Research Database (GPRD). Methods A convenience sample of 101 020 patients aged 40+ years drawn from GPRD formed the primary data source. This cohort was derived from a larger sample originally established for a cohort study of diabetes. GPRD records were linked with those from cancer registries in the National Cancer Data Repository (NCDR). Concordance between the two datasets was then evaluated. For cases recorded only on one dataset, validation was sought from other datasets (Hospital Episode Statistics and death registration) and by detailed analysis of a subset of GPRD records. Results A total of 5797 cancers (excluding non-melanomatous skin cancer) were recorded on GPRD. Of these cases, 4830 were also recorded on NCDR (concordance rate of 83.3%). Of the 976 cases recorded on GPRD but not on NCDR, 528 were present also in the hospital records or death certificates. Of the 341 cases recorded on NCDR but not on GPRD, 307 were recorded in these other two datasets. Rates of concordance varied by cancer type. Cancer registries recorded larger numbers of patients with lung, colorectal, and pancreatic cancers, whereas GPRD recorded more haematological cancers and melanomas. As expected, GPRD recorded significantly more non-melanomatous skin cancer. Concordance decreased with increasing age. Conclusion Although concordance levels were reasonably high, the findings from this study can be used to direct efforts for better recording in both datasets. Copyright © 2012 Crown copyright.

The effects of low persistence on fracture risk have not been fully addressed. The objectives of this study were to describe the persistence and compliance with bisphosphonates and to evaluate the association with fracture risk. The General Practice Research database was used to identify patients >or=18 yr of age prescribed alendronate or risedronate. The follow-up was divided into periods of current and past use. Time-dependent Cox regression was used. The study population included 44,531 patients; 58.3% of the patients continued bisphosphonate treatment for >1 yr and 23.6% for >5 yr. The risk of hip/femur fracture (adjusted relative rate [RR], 0.78; 95% CI, 0.64-0.94) and osteoporotic fracture (RR, 0.85; 95% CI, 0.76-0.94) were lower with current compared with past bisphosphonate use. The largest reduction in hip/femur and osteoporotic fracture risk was observed in patients treated for at least 6 mo and no reduction in those treated for

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT: * Use of beta(2) agonists has been associated with tachycardia, an abnormal ECG and atrial fibrillation. * Previous observational studies of the association between use of beta(2) agonists and the risk of acute myocardial infarction (MI) have demonstrated conflicting results. * Instead of a causal effect, the positive association between beta(2) agonist use and MI may be explained by latent ischaemic heart disease, which has symptoms that appear similar to respiratory complaints in chronic obstructive pulmonary disease.WHAT THIS STUDY ADDS: * The majority of beta(2) agonist users in our study population did not have an increased risk of nonfatal acute MI. * Only patients with ischaemic heart disease and who had recently started beta(2) agonists had an increased risk of acute MI. * It is likely that this increased risk was related to latent cardiovascular disease rather than direct effects of beta(2) agonists.AIM: Observational retrospective studies of the association between use of beta(2) agonists and the risk of acute myocardial infarction (MI) have demonstrated conflicting results, particularly among first-time users. The aim of this study was to examine the association between beta(2) agonist use and first nonfatal acute MI.METHODS: We conducted a case-control study (2476 cases) nested in a cohort of antihypertensive drug users in the Dutch PHARMO RLS database. PHARMO RLS consists of drug dispensing linked to the national hospitalizations register. Each case of nonfatal acute MI was matched with up to 12 control patients by gender, age and region. Drug and disease history and the severity of the underlying respiratory disease were adjusted for.RESULTS: Risk of acute MI was increased in current beta(2) agonist users [crude odds ratio (OR) 1.36, 95% confidence interval (CI) 1.15, 1.61]. However, this excess risk was reduced after adjustment for severity of asthma and chronic obstructive pulmonary disease (adjusted OR 1.18, 95% CI 0.93, 1.49). The risk was highest in patients with ischaemic heart disease and low cumulative dose of beta(2) agonists (adjusted OR 2.47, 95% CI 1.60, 3.82).CONCLUSION: Most users of beta(2) agonists did not have an increased risk of acute MI. Only patients with ischaemic heart disease with low cumulative exposure to beta(2) agonists had an increased risk of acute MI. It is likely that this increased risk was related to latent cardiovascular disease rather than to the direct effects of beta(2) agonists.

PURPOSE: To assess the association between exposure to antiepileptic drugs (AEDs) and the occurrence of aplastic anemia. METHODS: A retrospective case-control study was conducted using data from the U.K. General Practitioners Research Database (GPRD). Cases were defined as patients diagnosed with aplastic anemia. For each case, up to three control patients were matched on age, sex, and medical practice. Cases and controls were compared with respect to AED use. The effects of duration of AED use were assessed. Characteristics of individual cases with AED use were reviewed. RESULTS: The study population comprised 173 cases and 497 controls. AED use was more prevalent among cases (9.2%) than among controls (0.8%). After adjustment for confounders, the use of AEDs was significantly associated with aplastic anemia (adjusted odds ratio (OR), 9.5; 95% confidence interval (CI), 3.0-39.7). The most frequently used AEDs were carbamazepine (CBZ), valproic acid (VPA), and phenytoin. The 16 exposed cases were heterogeneous with respect to patient and exposure characteristics: the age of these patients varied from 1 to 92 years, and the duration of AED use varied from 17 days to 6.8 years. CONCLUSIONS: This study indicates that use of AEDs, in particular CBZ and VPA, is associated with a ninefold increased risk of aplastic anemia. Physicians should be alert to the possibility of AED-associated aplastic anemia.

Skeletal complications during or after treatment of acute lymphoblastic leukemia (ALL) have been frequently reported and can cause substantial morbidity, yet their incidence is not well established. The present study assessed the incidence of fractures, osteonecrosis (ON), and bone pain during ALL treatment and compared the fracture incidence with age- and sex-specific reference data from the UK General Practice Research Database (GPRD).Medical records of 122 ALL patients diagnosed at our institution from 1992 to 2004 were reviewed for information on fractures, ON, bone pain, and their anatomical location, risk group, phase of antileukemic therapy, and time since diagnosis. Evaluation of skeletal complications was followed up until July 2005 or the patient's death. Thirteen children were excluded as they were transferred to other institutions shortly after diagnosis.Skeletal complications occurred at a 5-year incidence of 32.7%. The 5-year incidence of fractures, ON, and isolated bone pain was 13.5%, 12.1%, and 12.3%, respectively. The relative rate of fractures adjusted for age and sex was 2.03 (95% confidence interval 1.15-3.57) compared to the GPRD, with greatest rates in children5 years. Thirty ON occurred in 10 patients with a 15 times greater incidence in children10 years than in those5 years. Nearly all skeletal complications occurred during maintenance therapy at a median of 14.92 months (range 0.0-53.8) after diagnosis and in weight-bearing bones.The doubled fracture rate and the high incidence of skeletal complications during the first years after diagnosis suggest the developing skeleton is very vulnerable in this period. Adolescents develop more ON whereas younger children may be more prone to fractures. Serious "immediate effects" of chemotherapy on bone appear of great concern and should entail preventative studies in this group of patients.

To compare the incidence of various fractures in a cohort of patients with epilepsy with a reference cohort of patients not having epilepsy.Patients were included in the epilepsy cohort if they had at least one diagnosis of epilepsy in their medical history and had sufficient evidence of "active" epilepsy (use of antiepileptic drugs, diagnoses) after the practice was included in the General Practice Research Database (GPRD). Two reference patients were sampled for each patient with epilepsy from the same practice. Primary outcome was the occurrence of any fracture during follow-up. Poisson regression analysis was used to estimate incidence density ratios (IDRs).The study population comprised 40,485 and 80,970 patients in the epilepsy and reference cohorts, respectively. The median duration of follow-up was approximately 3 years. The overall incidence rate in the epilepsy cohort was 241.9 per 10,000 person-years. This rate was about twice as high as that in reference cohort: age- and sex-adjusted IDR, 1.89 (95% CI, 1.81-1.98). When comparing IDRs among the different groups of fractures, the highest relative-risk estimate was found for hip and femur fractures (adjusted IDR, 2.79; 95% CI, 2.41-3.24). IDRs were consistently elevated across age and sex groups and across fracture subtypes.The overall risk of fractures was nearly twice as high among patients with epilepsy compared with the general population. The relative fracture risk was highest for hip and femur. Further study is necessary to elucidate whether this elevated risk is due to the disease, the use of antiepileptic drugs, or both.

Peterson & Naunton correctly observe that our study was not a randomized clinical trial (RCT) and that there is a potential for bias in observational studies. What are readers and the public to make of these general comments? Should we just discard observational evidence and wait for evidence to emerge from RCTs, as implied by Peterson & Naunton? In 1983, the renowned clinical epidemiologist Feinstein noted that ‘despite the magnificent scientific achievements of randomized clinical trials, the foundation for a basic science of patient care will also require major attention to the events and observations that occur in the ordinary circumstances of clinical practice’[1]. There are often substantial differences in the patients recruited into clinical trials and those in actual clinical practice. As an example, a recent analysis found that users of selective Cox-2 inhibitors were very different from patients participating in large RCTs. Most of these users in actual clinical practice would not have been eligible for inclusion into the pivotal RCTs, as they only used the medicines intermittently and for short periods of time [2]. There is insufficient RCT evidence for several of the outcomes evaluated in our study [3]. Of course, observational studies can yield incorrect answers, as highlighted by the controversies around hormone therapy. The challenge is to conduct an evidence-based assessment of bias rather than to merely speculate on potential biases. We tried to minimize the effects of bias by applying a variety of approaches, including an evaluation of patterns of risk. The risk factors used in the statistical analyses were selected a priori on the basis of likely association with outcome and ability to measure in our data source. The fit of statistical models was checked. Rigorously conducted observational studies are important as they reflect the real-life experiences with medication.