Epidemiology of immune thrombocytopenic purpura in the General Practice Research Database

Authors

Errata

This article is corrected by:

  1. Errata: Corrigendum Volume 147, Issue 1, 157, Article first published online: 14 September 2009

  • Funding source: Amgen, Inc.

W. Marieke Schoonen, Amgen Ltd., 1 Uxbridge Business Park, Sanderson Road, Uxbridge, Middlesex UB8 1DH, UK.
E-mail: mariekes@amgen.com

Summary

The epidemiology of immune thrombocytopenic purpura (ITP) is not well-characterised in the general population. This study described the incidence and survival of ITP using the UK population-based General Practice Research Database (GPRD). ITP patients first diagnosed in 1990–2005 were identified in the GPRD. Overall incidence rates (per 100 000 person-years) and rates by age, sex, and calendar periods were calculated. Survival analysis was conducted using the Kaplan-Meier and proportional hazard methods. A total of 1145 incident ITP patients were identified. The crude incidence was 3·9 (95% confidence interval [CI]: 3·7–4·1). Overall average incidence was statistically significantly higher in women (4·4, 95% CI: 4·1–4·7) compared to men (3·4; 95% CI: 3·1–3·7). Among men, incidence was bimodal with peaks among ages under 18 and between 75–84 years. The hazard ratio for death among ITP patients was 1·6 (95% CI: 1·3–1·9) compared to age- and sex-matched comparisons. During follow-up 139 cases died, of whom 75 had a computerised plausible cause of death. Death was related to bleeding in 13% and infection in 19% of these 75. In conclusion, ITP incidence varies with age and is higher in women than men. This potentially serious medical condition is associated with increased mortality in the UK.

Immune thrombocytopenic purpura (ITP) is an autoimmune disorder characterised by persistent thrombocytopenia (Cines & Blanchette, 2002, British Committee for Standards in Haematological General Haematology Task Force 2003; Cines & McMillan, 2005). Blood platelets in patients diagnosed with ITP undergo autoantibody destruction by phagocytosis, primarily in the spleen, leading to low platelet counts (Heyns Adu et al, 1986; Cines & Blanchette, 2002; Cines & McMillan, 2005). Impaired platelet production may also contribute to the low platelet counts (Stoll et al, 1985; Heyns Adu et al, 1986) through autoantibody inhibition of megakaryocyte function (Chang et al, 2003; McMillan et al, 2004) and altered levels of endogenous thrombopoietin (Kosugi et al, 1996; Nichol, 1998).

Thrombocytopenia is commonly defined by a threshold platelet count. Most studies use a threshold of 150 × 109/l (Cortelazzo et al, 1991; McMillan & Durette, 2004; Fogarty & Segal, 2007), but other values have also been used (e.g. Simpson et al, 1998; Frederiksen & Schmidt, 1999; Sutor et al, 2001; Neylon et al, 2003; Zeller et al, 2005). A normal platelet count ranges from 150 to 450 × 109/l (Kaushansky & Roth, 2004); the platelet counts of adult ITP patients are often <20 × 109/l at diagnosis. Symptoms of ITP and platelet count levels range widely among patients (British Committee for Standards in Haematological General Haematology Task Force 2003). In general, ITP patients with platelet counts of 50–150 × 109/l are asymptomatic and often diagnosed incidentally; platelet counts between 20 and 50 × 109/l are associated with excessive bruising with minor trauma; petechiae or ecchymoses develop spontaneously when counts are between 10 and 20 × 109/l and spontaneous bleeding may occur (Cines & Blanchette, 2002; Cines & Bussel, 2005). Patients with platelet counts below 10 × 109/l are at risk for severe bleeding, such as intracranial haemorrhage (ICH) or other internal bleeding (Cines & Blanchette, 2002). The risk of severe bleeding has been estimated to increase with increasing patient age and among patients with platelet counts persistently <30 × 109/l (Cohen et al, 2000).

The diagnosis of ITP is made by excluding other causes of thrombocytopenia, such as underlying infection with human immunodeficiency virus (HIV) or hepatitis C, autoimmune disease (e.g. systemic lupus erythematosus [SLE]), malignant lymphoproliferative and myeloproliferative disease and leukaemia, or use of specific medicines that may induce thrombocytopenia (e.g. heparin, alcohol, quinidine, quinine) (George et al, 1996, British Committee for Standards in Haematological General Haematology Task Force 2003; Cines & McMillan, 2005). The basic diagnostic approach to ITP includes a patient history, physical examination, complete blood count, and examination of a peripheral blood smear. Examination of the bone marrow is only recommended for patients aged over 60 years and in patients considering splenectomy (George et al, 1996).

Most patients with ITP respond to corticosteroid treatment with or without subsequent splenectomy and rarely need other immunosuppressive therapies (British Committee for Standards in Haematological General Haematology Task Force 2003), but a minority of ITP patients are refractory to all currently available treatments and have a persistently low platelet count (Stasi & Provan, 2004). These chronic refractory patients appear to be at the highest risk of severe bleeding and mortality (Cortelazzo et al, 1991; McMillan & Durette, 2004; Fogarty & Segal, 2007).

While the basic pathophysiology of ITP is understood, the epidemiology and clinical course of ITP have not been well investigated in the general population. The current literature includes a limited number of published studies, each of which describes a relatively small number of ITP patients. In the present study, we report up-to-date sex- and age-specific incidence estimates of ITP using the large United Kingdom (UK) population-based General Practice Research Database (GPRD). In addition, we describe the survival experience of ITP patients in the GPRD and compare their survival to that of patients without ITP.

Materials and methods

General practice research database (GPRD)

Since 1987, a varying number of general practices in the UK have contributed data to the GPRD. The number of actively registered patients covered by participating practices has increased from roughly 1 million in 1990 to over 3·4 million in 2005 (i.e. 2–6% of the UK population) (General Practice Research Database 2008). General practices and individuals included in the GPRD are representative of the UK as a whole (Hollowell, 1997). The GPRD is the largest population-based primary care database available for research to date.

Data in the GPRD are recorded by practice staff and include details of patient registration, detailed prescription information, medical information recorded through the Oxford medical information system (OXMIS) and Read codes for symptoms, diagnoses and laboratory tests (Lis & Mann, 1995; Robinson et al, 1997). The database also contains information on referrals to specialists and hospital discharge letters, including date, type of specialist seen, diagnosis and presenting symptoms (van Staa & Abenheim, 1994). Information is recorded on a patient level using an anonymised patient identifier (Hollowell, 1997). Data from individual practices undergo a series of quality checks before they are uploaded into the centralised database; practices with inconsistent or incomplete data are not included and practices passing the checks are considered to be ‘up-to-standard.’ The version of the GPRD held by the Boston Collaborative Drug Surveillance Program (‘Boston GPRD’) contains data from practices that have passed additional quality checks and are known to provide copies of medical records on request (Jick et al, 2003). Studies investigating the data quality in this version of the GPRD have indicated high completeness and validity (Jick et al, 1991, 2003). Data from the Boston GPRD were used in the current study. Studies comparing the information on copies of referral and discharge letters to the computerised information contained in the GPRD have found a high positive predictive value and sensitivity for a wide range of diagnoses (van Staa & Abenheim, 1994; Jick et al, 2003).

Study population

The base population for this study consisted of all individuals who were actively registered with one of the approximately 350 practices that contributed data to the Boston GPRD in 1990–2005. Each patient’s GPRD entry date was the date of their initial registration or the date their practice was judged to be contributing up-to-standard data, whichever was later. The date of last follow-up in the GPRD was the date the patient died or transferred out of their practice, or the date of the last data upload to the GPRD from their practice, whichever was earliest.

ITP patients

We identified all patients with at least one diagnostic code for ITP in their medical records. Medical codes denoting a diagnosis of ITP are listed in Table I. The first mention of an ITP medical code was defined as the ITP diagnosis date (index date). If this date was during the period 1990–2005, the patient was considered an incident ITP patient in this study. Patients with a first mention of ITP before the start of the study period were considered prevalent rather than incident, and were not included in the analyses presented here. For patients who died during the study period, date of death was defined as the date that a clinical diagnosis of death was recorded by the general practitioner (GP).

Table I.   Codes from the GPRD coding dictionary used to identify ITP patients and platelet counts.
CodeDescriptionClassification
  1. Nos, not otherwise specified.

42P2·11Auto-immune thrombocytopeniaRead
D313·11Evan’s Syndrome Read
D313·12Idiopathic Thrombocytopenic PurpuraRead
D313000Idiopathic Thrombocytopenic PurpuraRead
D313012ITP–Idiopathic Thrombocytopenic PurpuraRead
2871CIdiopathic ThrombocytopeniaOXMIS
L 146NPlatelet countOXMIS
42P..00Platelet countRead
42PZ.00Platelet count, nosRead

We investigated the clinical characteristics of all incident ITP patients in further detail, in terms of their platelet counts and cause of death for patients who died during the study period. Platelet counts were identified from computerised records of laboratory test results (for OXMIS and Read codes denoting a laboratory test see Table I). Among patients for whom at least one results of a platelet count test was recorded, we used the lowest platelet count recorded at any time for each patient. Additionally, we identified the number of patients with a splenectomy recorded in their medical files on or after the ITP diagnosis date.

Among incident ITP patients who died during the study period, cause of death was investigated by manually reviewing their computerised medical records to identify diagnoses listed at the time of death. Diagnoses listed within one week of the clinical diagnosis of death were considered possible causes of death. These were grouped into diagnoses that were potentially related to bleeding and diagnoses of thrombocytopenia or ITP listed in association with a diagnosis of death with no specific mention of bleeding.

To investigate the accuracy of computerised ITP diagnosis in the GPRD, we randomly sampled 150 patients who were alive and actively registered with their general practice at the end of the study period. Their practices were requested to provide copies of paper medical records (including consultant letters and hospital discharge summaries), which were used to validate the computerised diagnosis of ITP. If evidence of a diagnosis of ITP was documented in the copied medical records, the computerised diagnosis was considered as confirmed. The computerised diagnosis was considered unconfirmed if (i) insufficient information was available from the original records (e.g. a single mention of the word ‘thrombocytopenia’) or (ii) if a disease causing thrombocytopenia (e.g. lymphoproliferative disease) was diagnosed after ITP and the records were judged to better support the alternative diagnosis.

In addition to investigating the accuracy of a computerised ITP diagnosis compared to patient’s original medical records, we also explored the accuracy of ITP diagnoses as established by GPs. Because a diagnosis of ITP is made by excluding other causes of thrombocytopenia, potentially false positive diagnoses of ITP are those where ‘other causes’ (i.e. medical conditions potentially related to thrombocytopenia or drug prescriptions potentially inducing thrombocytopenia), are listed concurrently in a patient’s computerised medical file. A list of comorbid medical conditions and drug prescriptions of interest are listed in Table II (Liebman, 2007). If an ITP patient was diagnosed with any of these conditions before or within six months after the ITP diagnosis date, then this case was considered a potentially false-positive case of ITP. To identify individuals with drug-induced thrombocytopenia who were potentially misclassified by their GP as ITP patients, we searched the computerised therapy records for prescriptions for heparin, quinine, quinidine and cytotoxic chemotherapy issued within 60 days prior to a patient’s ITP diagnosis date. Administration of chemotherapy is not recorded in the GPRD as this treatment is given outside of primary care. To explore how many patients with chemotherapy-induced thromobocytopenia may have been incorrectly diagnosed as ITP, we determined how many patients had a diagnosis of a malignancy in the year before the ITP diagnosis to reflect the time period during which active cancer treatment, including chemotherapy, may have been received.

Table II.   Demographic characteristics of incident patients with ITP diagnosed in 1990–2005.
 Study subjects (%)Person-years (million)
  1. *Diagnosed before or within 6 months after ITP diagnosis.

  2. †Diagnosed in the year before ITP diagnosis.

  3. ‡Drug prescription recorded within 60 days prior to ITP diagnosis date.

Total number ITP patients114529·2
Diagnosis year by calendar period
 1990–1994187 (16·3)6·1
 1995–1999312 (27·2)9·3
 2000–2005646 (56·4)13·8
Age at diagnosis  
 Under 18257 (22·4)6·1
 18–64534 (46·6)18·3
 65–100354 (30·9)4·8
Sex  
 Female652 (56·9)14·8
 Male493 (43·1)14·4
Co-morbid conditions*100 (8·7) 
 Haematologic malignancy39 (3·4) 
 Systemic lupus erythematosus15 (1·3) 
 Evan’s syndrome8 (0·7) 
 Non-haematologic malignancy†8 (0·7) 
 Hepatitis B virus7 (0·6) 
 Auto-immune haemolytic anaemia6 (0·5) 
 Anti-phospholipid antibody syndrome5 (0·4) 
 Human immunodeficiency virus3 (0·3) 
 Myelodysplastic syndrome3 (0·3) 
 Hepatitis C virus2 (0·2) 
 Haemolytic uremic syndrome0 
 Thrombotic thrombocytopenic purpura0 
 Disseminated intravascular coagulation0 
 Drug induced thrombocytopenia0 
 Congenital thrombocytopenia0 
Concomitant medications‡  
 Heparin0 
 Quinidine 12 (1·0) 
 Quinine0 

Comparison subjects without ITP

Up to five individuals without a recorded diagnosis of ITP were randomly selected for each incident ITP patient. The comparison subjects were individually matched by age, sex, and calendar period and, where possible, by practice. If insufficient comparison subjects were available within an ITP patient’s practice, they were randomly selected from outside the practice. The index date of comparison subjects was defined as the date of ITP diagnosis of the matched ITP patient. Observation times of comparison subjects who were diagnosed with ITP after their assigned index date were censored on the date of their first diagnosis of ITP.

Statistical analyses

Incidence

The overall incidence of ITP was estimated as the number of patients with a first-time recorded diagnosis of ITP each year divided by the person-time contributed by all subjects in the database without ITP that year. Person-time for individuals who developed ITP, died, or transferred out of their practice during a given year was censored at the time of the event. Ages were grouped as follows (in years): under 18, 18–24, 25–34, 35–44, 45–54, 55–59, 60–64, 65–74, 75–84, and 85–100. We also created three broader age categories to represent children, persons of working age, and persons of retirement age (i.e. under 18 years, 18–64 years, and 65–100 years). Subjects with recorded ages of more than 100 years at the time of ITP diagnosis were excluded from analysis. Calendar time was divided into the following five-year periods: 1990–94, 1995–99, and 2000–05. Ninety-five per cent confidence intervals (CIs) were estimated using Byar’s method (Rothman & Boice, 1979).

Survival

The survival functions of patients with ITP and their matched comparison subjects were estimated separately using the Kaplan–Meier product limit method (Kaplan & Meier, 1958). Survival difference between the groups was tested using the log-rank procedure (Kalbfleisch & Prentice, 1980). We estimated hazard ratios for death comparing the ITP patients to the comparison group using Cox regression (Cox & Oakes, 1984).

Validation of the computerised ITP diagnosis

We established the proportion of false positive ITP diagnoses (i.e. the number of patients for whom the ITP diagnosis was not confirmed after review of medical records, divided by the total number of patients for whom copies of medical records were received) and the positive predictive value of a computerised ITP diagnosis (i.e. the number of patients with a confirmed ITP diagnosis divided by the total number of ITP patients identified by means of their computerised information).

We computed frequencies of patient characteristics and causes of death. Statistical significance was considered at a P-value of <0·05. Data management and statistical analyses were carried out using sas version 9.1 (SAS Institute, Cary, NC, USA) and stata version 8.2 (StataCorp, College Station, TX, USA).

Results

We identified 1145 patients with incident ITP among 29·2 million person-years of observation during 1990–2005. The characteristics of these patients are listed in Table II. The 1145 incident ITP patients were identified using six distinct diagnosis codes (Table I), with 1044 (91·2%) having one of the diagnostic codes for ITP, 92 (8·0%) having two different diagnostic codes for ITP recorded, and 9 (0·8%) having three different diagnostic codes for ITP. Data on the number of incident ITP patients identified and the person-time of follow-up are displayed in Table II. Men and women contributed a similar amount of person-years of follow-up during the study period (14·4 million and 14·8 million, respectively). The majority (62·7%) of person-years were contributed by individuals of the 18–64 year age group and approximately 47·3% of all person-years were contributed during the calendar period 2000–05. More ITP patients were diagnosed among adults aged 18–64 years [46·6%]) than among children (under age 18 years) and those aged over 65 years (22·4% and 30·9%, respectively). The majority of incident ITP patients were female (652 patients, 56·9%) compared to 493 (43·1%) male patients. The number of incident ITP patients with medical conditions potentially related to thrombocytopenia was low (Table II).

Incidence

The crude incidence of ITP for the 15-year study period was 3·9 per 100 000 person-years (95% CI: 3·7–4·1) (Table III). The overall average incidence rate for women (4·4 per 100 000 person-years; 95% CI: 4·1–4·7) was statistically significantly higher than for men (3·4 per 100 000 person-years; 95% CI: 3·1–3·7) (Table III). The age-specific incidence of ITP appeared to have a bimodal distribution for men (Fig 1), with the first peak incidence observed among boys (under 18 years old) and the second peak among men aged 75–84 years. For women, incidence rates were observed to be constant from childhood until approximately 60 years old, after which the incidence increased with increasing age. A higher incidence rate was observed for both men and women in the period 2000–05 compared to the periods 1990–94 and 1995–99.

Table III.   Overall and sex-specific incidence rates per 100 000 person-years of observation time of incident ITP by calendar period and age.
 MaleFemaleOverall
NIncidence (95% CI)NIncidence (95% CI)NIncidence (95% CI)
Overall4933·4 (3·1–3·7)6524·4 (4·1–4·7)11453·9 (3·7–4·1)
Calendar period
 1990–94 762·5 (2·0–3·1)1113·6 (2·9–4·2)1873·1 (2·6–3·5)
 1995–991212·6 (2·2–3·1)1914·0 (3·5–4·6)3123·3 (3·0–3·7)
 2000–052964·4 (3·9–4·8)3505·0 (4·5–5·5)6464·7 (4·3–5·1)
Age group (years)
 Under 181484·7 (3·9–5·5)1093·7 (3·0–4·4)2574·2 (3·7–4·7)
 18–641882·0 (1·7–2·3)3463·8 (3·4–4·2)5342·9 (2·7–3·2)
 65–1001577·8 (6·6–9·0)1977·1 (6·1–8·0)3547·4 (6·6–8·1)
All adults >183453·1 (2·7–3·4)5434·6 (4·2–5·0)8883·8 (3·6–4·1)
Figure 1.

 Average annual ITP incidence by age group and gender (n = 1145).

Clinical characteristics of incident ITP patients

Among the 1145 incident ITP patients, 747 (65·2%) had at least one recorded platelet count. The proportion of incident patients with at least one recorded platelet count was higher in later years of the study period (approximately 30% of those diagnosed in 1990 increasing to nearly 80% of patients diagnosed in 2005, data not shown). A total of 593 of the 747 patients (79·4%) with at least one recorded platelet count were observed to have a lowest recorded count of <150 × 109/l; 65·9% of patients had a count of <100 × 109/l. The frequency distribution of lowest recorded platelet counts is provided in Fig 2.

Figure 2.

 Distribution of lowest platelet count recorded among incident ITP patients (n = 747). Each bar represents an interval of 10 × 109 platelets/l except for the bar at the far right, which represents all patients with a lowest recorded platelet count of 150 × 109/l or greater.

Prescription records of the 1145 incident ITP patients revealed that 396 (34·6%) received at least one prescription for prednisolone (the oral corticosteroid used nearly exclusively in the UK) and 66 patients (5·8%) had at least one prescription for azathioprine. A total of 70 ITP patients (6·1%) underwent splenectomy on or after their index date. A splenectomy was found to be listed for 44 (3·8%) of the ITP patients at any time prior to their first recorded diagnosis of ITP in the GPRD.

Cause of death

One hundred and thirty-nine (12·1%) patients died during their follow-up time recorded in the GPRD up to 2005, 75 of whom had plausible causes of death identified in their records. Ten of these 75 patients (13% of patients with a plausible cause of death listed) had causes of death related to bleeding, including single patients with melena, gastrointestinal haemorrhage or duodenal ulcer, haematuria, haemorrhage not otherwise specified, intraoperative haemorrhage, postoperative haemorrhage, and rectal bleeding and three patients with intracranial haemorrhage (one labelled intracranial haemorrhage and two labelled intracerebral haemorrhage). Four of the ten patients with a cause of death related to bleeding also had low platelet counts (1–141 000 × 109/l) documented in the GPRD within a month prior to death. Three additional cases of the 75 with plausible causes of death recorded had diagnoses of thrombocytopenia or ITP listed in association with the clinical diagnosis of death, although no specific bleeding-related cause of death was recorded. One of these three patients died of a pulmonary infarction, which was probably not related to bleeding. The other two died of a stroke and a cerebrovascular accident, with insufficient data to confirm whether these were haemorrhagic strokes. Fourteen patients (19% of patients with a plausible cause of death listed) had an infectious disease recorded as the primary cause of death. In the remaining 48 ITP patients who died, sufficient evidence was not found to enable the conclusion that causes of death were directly linked to bleeding or infection.

Survival of incident ITP patients and comparison subjects

We identified 5702 age- and sex-matched comparison subjects for the 1145 incident ITP patients. Of these, 139 ITP patients (12·1%) and 469 comparison subjects (8·2%) died during follow-up upto 2005 (Table IV). The median follow-up time for the incident ITP patients as well as the comparison subjects was 3·4 years (75th percentile: 6 years, with maximum follow-up of 15 years for cases and 15·1 years for comparison subjects). A log-rank test comparing the proportion of deaths among ITP patients versus the proportion among comparison subjects, stratified by follow-up time, showed a statistically significant difference (= 0·0001). The estimated age- and sex-adjusted hazard ratio of death among ITP patients compared to age- and sex matched comparison subjects was 1·6 (95% CI: 1·3–1·9).

Table IV.   Number of deaths among patients with incident ITP and comparison subjects.
Age group (years)Deaths among ITP patients [N deaths/N total, (%)]Deaths among comparison subjects [N deaths/N total, (%)]
<182/257 (0·8)1/1285 (0·1)
18–6422/534 (4·1)45/2670 (1·7)
65–100115/354 (32·5)423/1747 (24·2)
Overall139/1145 (12·1)469/5702 (8·2)

Of the 139 ITP patients who died, 79 (56·8%) were female and 60 (43·2%) were male. The risk of mortality, adjusted for age and disease status, was not significantly different when comparing men and women (log-rank test = 0·07). Few deaths (<2%) were observed among incident ITP patients and comparison subjects who were younger than 45 years of age (Table IV). A greater proportion of deaths was observed among ITP patients (32·4%) aged 65 years and older than among the comparison group (24·2%).

Validation of the computerised ITP diagnosis

Of the 150 randomly selected patients for whom copies of medical records were requested, we received a response from the GP for 115. No additional medical records (such as consultant letters or hospital discharge summaries) were available for ten of the 115 patients. GPRD staff anonymised copies of the remaining 105 patient records, of which we received information on 102. The diagnosis of ITP was confirmed in 93 (91%) of these 102 patients; i.e. the positive predictive value of a computerised diagnosis of ITP was estimated to be 91% (95% CI: 84–96%). For nine patients, the ITP diagnosis could not be confirmed for the following reasons: insufficient information in the available medical records (three patients), purpuric rash associated with viral infection (two patients), diagnosis with known potential aetiology for thrombocytopenia (acute myeloid leukaemia [one patient], myelodysplastic syndrome [one patient], sepsis [one patient]), and incorrect diagnosis recorded on computer when the true diagnosis was essential thrombocythaemia rather than thrombocytopenia (one patient).

Discussion

This large, population-based study found that the incidence of ITP in British women is higher compared to men, particularly in the 18–64 year age group (crude incidence of 3·8 vs. 2·0 per 100 000, respectively). We found increased mortality among incident ITP patients compared to the general population, with an estimated age- and sex-adjusted hazard ratio of death of 1·6 (95% CI: 1·3–1·9). A cause of death related to bleeding was found in 10 ITP patients and a further 14 patients were observed to have an infection as cause of death (13% and 19% of 75 ITP patients with recorded causes of death, respectively).

Few studies published in the peer-reviewed scientific literature have presented descriptive epidemiology of ITP based on population-based data. With the inclusion of more than 1000 incident patients with ITP, the current study is the largest reported to date. Our finding of an overall incidence of 3·9 patients per 100 000 person-years is higher than the estimates of 1·6 in England (Neylon et al, 2003) and 2·68 in Denmark (Frederiksen & Schmidt, 1999). Feudjo-Tepie and Logie (2007) utilised the GRPD to estimate incidence of adult ITP in the years 1992–2005. They found 840 cases, resulting in an adult ITP incidence rate of 3·9 per 100 000 person-years. When restricting our data to adult incident cases of ITP (i.e. 18 and older), we found an overall incidence rate of 3·8 per 100 000. The discrepancy between our estimate and the estimate of Neylon et al (2003) may be related to their use of a relatively low platelet count threshold (<50 × 109/l) for the diagnosis of immune thrombocytopenia, and restricting their ITP patient population to individuals who had undergone bone marrow examination to exclude other haematological pathology. The difference between our estimate and that of Frederiksen and Schmidt (1999) may reflect actual differences between the populations of the United Kingdom and Denmark, but may also be related to the time periods considered. The incidence during the first half of their study (from 1973 to 1984) was 1·94 per 100 000. During the latter half of the Danish study (from 1985 to 1995) the incidence was 3·33 per 100 000 person-years (Frederiksen & Schmidt, 1999), which is similar to our overall incidence estimate (using UK data from 1990 to 2005). Similar to the Danish study, we also observed an increase in ITP incidence over time with a higher incidence after the year 2000. This increase in incidence cannot be explained by an increase in population size because incidence estimates take into account the person-time at risk of the study population. The observed increase may be due to changing coding practices in the GPRD by physicians, with the switch to new software around the year 2000 and may thus represent an artefact in the data. Alternatively the observed increase in ITP incidence may be due to increased awareness of the disorder among physicians, or an unexplained increase in exposure to risk factors for ITP. The data reported here do not allow us to distinguish among these possibilities.

The observed variation in ITP incidence by age in this study was similar to that of previous studies. Neylon et al (2003) and Frederiksen and Schmidt (1999) both estimated a higher incidence of ITP among those aged 60 years and older. The higher estimated incidence reported in the 65- to 100-year-old age group may be related to a truly higher incidence or to an increased risk of bleeding as a presenting symptom leading to a diagnostic workup. In a group of 117 adult ITP patients, severe bleeding events were observed when platelet counts were below 30 × 109/l, and the risk of bleeding was strongly increased among those over age 60 years (versus patients younger than 40 years) (Cortelazzo et al, 1991). The higher incidence may also be related to the increased number of GP encounters older patients may have for other age-related symptoms and diseases, leading to an increased likelihood of ITP being diagnosed.

For adults, previous studies have indicated a higher incidence among females compared to men, with an attenuation of this difference in older age groups (Frederiksen & Schmidt, 1999; Neylon et al, 2003). Among all adults older than 18 years, we observed a difference by sex with an incidence for men of 3·1 per 100 000 person-years, compared to 4·6 per 100 000 person-years for women. When stratifying by age group, women had a nearly twofold higher incidence than men in the age group of 18- to 64-year-olds, and this shifted to men having a slightly higher incidence than women in the 65- to 100-year-old age group.

A diagnosis of ITP is made by excluding other causes of disease or use of specific medicines that may induce thrombocytopenia (Cines & McMillan, 2005). An estimated 1% of ITP patients also have autoimmune haemolytic anaemia (Cines & McMillan, 2005), a syndrome also known as Evans’ syndrome. In our patient population, the records of 10 patients (3 female, 7 male) out of 1145 incident ITP patients also contained a code for Evans’ syndrome (0·9%), the majority of whom were patients with an ITP diagnosis date after the year 2000. Overall, 100 patients (8·7%) had at least one comorbid condition or prescription for a medication potentially related to development of thrombocytopenia during a relevant time period. When these 100 patients were omitted from our calculations, the overall incidence rate decreased slightly, from 3·9 to 3·6 per 100 000 person-years (95% CI: 3·4–3·8) (data not shown). From these data it appears that for the vast majority of cases who have a code for idiopathic thrombocytopenia, the GP correctly utilized this code since there was no evidence for alternative causes of disease among 91·3% of the cases. However, our estimates of the frequency of comorbid conditions among ITP patients should be interpreted with caution as we were not able to prospectively and systematically capture information on some relevant concomitant illnesses, as can be done in case series of ITP patients diagnosed at referral centres. For example, we did not include diagnoses of ‘viral hepatitis’ or ‘infective hepatitis’ in our search algorithm for concomitant hepatitis B or C as these diagnoses are non-specific, so we would have also captured cases without specific conditions predisposing to ITP. In addition, HIV infections may not have been reported in the patient’s computerised medical record, to protect the patient’s privacy or because these patients were diagnosed in sexual health clinics which do not necessarily communicate with a patient’s GP. This may partially explain why in our study fewer ITP patients (9%) appear to have secondary causes of thrombocytopenia than the proportion (28%) observed by Portielje et al (2001). In their study, 213 consecutive patients at a referral centre were evaluated systematically and treated according to a well-defined protocol. Neylon et al (2003) found that 40 cases out of their series of 245 (16%) may have had secondary thrombocytopenia. However, 18 of these 245 cases (7%) had a potential cause other than infection, which is in line with our findings.

Forty-four patients had evidence of undergoing a splenectomy before the date of ITP diagnosis and may have had the onset of ITP before their first recorded code for ITP in their computerised GPRD records (i.e. they were in fact prevalent rather than incident cases). Omitting these patients in the incidence calculations did not appreciably alter the results (data not shown).

Validation of a computerised diagnosis of ITP compared to copies of paper medical records indicated a high positive predictive value. Thus, when using medical codes alone from the GPRD to identify patients with ITP, approximately 9 in 10 of the patients identified were expected to truly have a diagnosis of ITP. The validation study did not investigate if any patients with non-specific recorded diagnoses of thrombocytopenia or purpura who in fact had ITP, were misclassified by their GP (i.e. patients with a false-negative diagnosis of ITP). Including such a wider range of diagnostic codes in searching for ITP cases in the GPRD would probably have included some additional true ITP cases, but would have increased the number of false positive diagnoses in our study by a much larger proportion and we would have therefore been likely to incorrectly overestimate the incidence of ITP.

The validation study did not include ITP cases who died or transferred out of their practice because practices do not retain records for such cases. We observed a difference in median age of the patients assessed for validation versus those not assessed (37·5 and 48 years, respectively) which was statistically significant (P = 0·02). This difference is probably a result of the fact that the risk of death among patients with ITP increases with increasing age. Since the median age of patients assessed for validation was lower than of the patients not assessed, it is possible that patients who were included in the validation sample may represent patients with less severe ITP than the overall population of ITP cases. Although it is plausible this may have affected the positive predictive value it is not clear whether it resulted in an under- or overestimate.

Review of the platelet counts recorded in the GPRD revealed that almost 80% of the patients had a lowest recorded count of <150 × 109/l. The lowest recorded platelet counts were <50 × 109/l for 333 patients (45% of 747 with at least one count recorded) and <25 × 109/l for 219 patients (37%). Platelet counts in the latter range are clinically relevant as these are often associated with petechiae, spontaneous or excessive ecchymoses, and clinically important bleeding (Cines & Blanchette, 2002). A combination of data on both diagnostic codes and platelet counts may increase the specificity of a computer-recorded ITP diagnosis in future studies in the GPRD, especially if the frequency with which platelet count test results are recorded continues to increase over time.

In our study we were unable to analyse the use of certain treatments for ITP that are typically administered in secondary care, such as intravenous immune globulins, other antibody-based treatments, platelet transfusions and other treatments prescribed by specialist consultants, because treatments prescribed outside the primary care setting are not systematically recorded in the GPRD. This was confirmed by our further evaluation of the medical records of the 102 patients who were assessed for diagnostic validation, where we found that eleven patients (11%) had received treatment with intravenous immune globulin but a computerised prescription code for treatment with intravenous immune globulin was not recorded in their GPRD records. By contrast, use of corticosteroids is likely to be captured relatively accurately and completely in the GPRD. Although in some instances corticosteroid therapy may have been initiated by a specialist or in hospital, repeat prescriptions for oral medications are frequently issued by GPs in the UK. Therefore the proportion of patients treated with oral prednisolone is not likely to be a gross underestimate of the true treated fraction of ITP patients.

Patients with persistently low platelet counts may be treated with splenectomy. We observed a lower proportion of patients who underwent splenectomy after their ITP diagnosis date (6·1%) than Neylon et al (2003), who found that 12% of patients eventually underwent splenectomy. It is thought that the frequency of splenectomy has declined over the last 30 years, partly due to physicians delaying the procedure since ITP has the potential for late remission (Godeau et al, 2007). Our study included more recent data than the Neylon study and our lower proportion of splenectomised ITP patients may be partly explained by this changing viewpoint on splenectomy. Other reasons for the observed differences in splenectomy rates between the two studies are differences in disease severity due to a lower platelet count threshold for ITP diagnosis used in Neylon et al’s study, and a different duration of follow-up available for observation.

Mortality in this study was approximately 60% higher among incident ITP patients than among matched comparison subjects. This mortality difference occurred predominantly among older subjects (aged 45 years and older). An increase in mortality among patients with ITP could be due to complications of ITP itself, to underlying medical conditions associated with ITP, or to treatments for ITP (e.g. infection due to immunosuppression). We did not adjust mortality estimates for the existence of co-morbid disease states such as cardiovascular disease. Of the 75 ITP patients for whom a cause of death was recorded, 13% likely died due to bleeding, which is a similar proportion as reported previously (Neylon et al, 2003). Interestingly, we found that the cause of death was of infectious origin in 19% of patients who died with a known cause of death listed in their records. This finding is in accordance with the findings of Portielje et al (2001) who observed the same percentages in a case series of 152 patients with a median follow-up of 10·5 years.

The GPRD coding system does not contain codes for ITP remission or relapse; therefore, we were unable to reliably distinguish patients with acute ITP from patients with chronic ITP. Chronic ITP is generally reported to be more frequent in adults than in children, and some evidence indicates that the proportion of chronic ITP increases not only from childhood to adulthood, but also from infancy to adolescence (Kuhne et al, 2001, Kuhne, et al 2003; Lowe & Buchanan, 2002). Although the clinical course of disease among paediatric ITP patients may be different from that among adult patients (e.g. children with acutely presenting symptoms may be hospital-based), children will be seen by a specialist based on a referral letter from the GP. As hospital referral and discharge letters are captured in the GPRD (van Staa & Abenheim, 1994; Jick et al, 2003) it is unlikely that our estimate of the ITP incidence among children is an underestimate.

Our report provides a robust population-based estimate of the incidence and survival of ITP in the UK and confirms that the incidence of ITP increases with age and is higher among women than men. To our knowledge, no other studies to date have investigated the mortality among ITP patients compared to the general population. We observed a 60% increased mortality and found evidence of bleeding or infection as the cause of death in a substantial number of ITP patients, illustrating that ITP is a serious and potentially life-threatening illness.

Conflicts of interest disclosure

This study was funded by Amgen, Inc., the manufacturer of Nplate.

Acknowledgements

We wish to thank Drs. Hershel and Susan Jick for helpful discussions and the general practitioners who contribute information to the GPRD for their ongoing efforts.

Author’s contributions

Investigators at the Boston Collaborative Drug Surveillance Program (BCDSP) designed this study in collaboration with the sponsor. BCDSP investigators carried out the work related to data extraction from the GPRD and analysis of GPRD data. Authors at Amgen and Exponent drafted the manuscript based on their review of the epidemiologic literature and study reports provided by the BCDSP investigators. Exponent authors also provided additional calculations of incidence stratified by age. Dr Rutstein provided clinical guidance and input on disease and diagnostic definitions, and critically reviewed drafts of the manuscript. All authors had responsibility for revising the manuscript and approved its final contents.

Ancillary