Thrombotic thrombocytopenic purpura is a rare disorder characterized by platelet aggregation and microthrombi leading to depletion of circulating platelets. The classic features of thrombotic thrombocytopenia purpura are thrombocytopenia, microangiopathic hemolytic anemia, neurologic symptoms and signs, renal abnormalities, and fever. Hemolytic uremic syndrome shares many of the features of thrombotic thrombocytopenia purpura except neurologic manifestations.1
Ischemia resulting from platelet microthrombi is believed to be central in the pathogenesis of thrombotic thrombocytopenia purpura and hemolytic uremic syndrome. In acquired thrombotic thrombocytopenia purpura, this results from accumulation of ultralarge multimers of von Willebrand factor, usually the result of inhibition of von Willebrand factor cleaving metalloproteinase (ADAMTS13) by antibodies.2–4 In hemolytic uremic syndrome, by contrast, the activity of ADAMTS13 is usually normal and ultralarge von Willebrand factor is not usually increased.4
Several factors have been implicated in the etiology of thrombotic thrombocytopenia purpura and hemolytic uremic syndrome, including vascular procedures, infections, chemotherapeutic agents, immunosuppressants, lipid-lowering agents, and some antiplatelet drugs.1 Supporting data for many risk factors are circumstantial, based on case reports and case series.5 Ticlopidine, an antiplatelet medication, is the only drug for which a quantitative increase in risk has been found,6 reportedly mediated by antibody formed against ADAMTS13.7 Similar concern has been raised for a related antiplatelet agent, clopidogrel, but on the basis of less direct information. In a case series based on intensive ascertainment, 11 thrombotic patients with thrombocytopenia purpura exposed to clopidogrel were identified,8 but the absence of information on the population base made the finding difficult to interpret quantitatively.
Historically, the case-fatality rate for untreated thrombotic thrombocytopenia purpura approached 100%, but today it is 20% or less with intensive plasmapheresis.9 Based on death certificates and an assumed case-fatality rate of 70%, Török10 estimated that the incidence of thrombotic thrombocytopenia purpura rose in the United States from 0.4 per million per year to 1.1 during 1970 to 1991. These estimates were necessarily indirect and, moreover, might not describe the current epidemiology of thrombotic thrombocytopenia purpura.
A more recent estimate of hemolytic uremic syndrome in California in children under 5 years of age showed no major change in the incidence rate between 1994 and 1999. The average incidence rate was 0.67 per 100,000 children per year.11
We evaluated the incidence of thrombotic thrombocytopenia purpura and hemolytic uremic syndrome in 3 large populations in the United States, the United Kingdom, and Canada. We conducted a case-control study to evaluate potential risk factors for thrombotic thrombocytopenia purpura and hemolytic uremic syndrome. We used automated healthcare records and accessed underlying medical files to obtain a more accurate set of incidence estimates, against which newer reports of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome in association with drugs and procedures can be compared.
We drew information from 1) the Ingenix Research Database in the United States; 2) the United Kingdom General Practice Research Database (GPRD); and 3) the Research Database from the Province of Saskatchewan. The first is based on health insurance claims submitted to a large, national health insurer in the United States, covering 18 states for the current study. The second is a database built on computerized medical records of general practitioners in the United Kingdom who use a practice management system. The third is an administrative database used in the delivery of a universal publicly funded health insurance program in the province of Saskatchewan.
The holders of each of these databases facilitate access to the underlying medical record, through appropriate intermediaries, and with complete anonymity of extracted data.
For the United States, the observation period began either on January 1, 1990, or the day after the end of a 3-month baseline period of health plan membership, whichever was later. Observation continued until either December 31, 2000, or termination of membership, whichever came first. To estimate the total population observation time covered by the insurance claims data, we selected records corresponding to a 1% random sample of the membership and reviewed eligibility status month by month. For each record in the 1% sample and each calendar month of the study period, we checked the 3-month continuous enrollment eligibility criterion as of the 15th day of the month. The eligible months of the sampled records were then summed in categories of age, sex, and calendar year. The sums were multiplied by 100 and divided by 12 to produce estimates of the person-years represented by the claims records. These estimates, based on over 80,000 records, served as the denominator for the incidence rate calculations.
For GPRD, the whole population in active practices used by the Boston Collaborative Drug Surveillance Program (BCDSP) was examined to calculate the population time under observation as the denominator for the incidence rate calculations. No criteria for minimum period of observation were imposed, because these records contain key elements of medical history. Patients were continuously enrolled from their registration date until they transferred out of practice or died, or until the patient's practice provided its last data update. The study period for the GPRD database was the same as for the Ingenix database.
The study period for the Saskatchewan cohort was from January 1, 1976, through December 31, 1999. Pharmacy data between July 1, 1987, and December 31, 1988, were not available. The province supplied denominator data.
Criteria for Case Identification
We identified thrombotic thrombocytopenia purpura and hemolytic uremic syndrome cases by review of the claims and computerized medical record data, followed by verification through review of the medical records.
The criteria used to identify potential cases in the 3 populations were any of the following:
- More than one recorded diagnosis of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome by a physician;
- Physician diagnoses of microangiopathic hemolytic anemia and thrombocytopenia during the same hospitalization or within 15 days of one another in an outpatient setting;
- Physician diagnoses of acquired hemolytic anemia and thrombocytopenia during the same hospitalization or within 15 days of one another in an outpatient setting;
- Diagnosis of either nonautoimmune acquired hemolytic anemia or thrombocytopenia within the 15 days before or after a plasma exchange procedure, or a therapeutic plasmapheresis procedure.
We extracted medical and pharmacy electronic data for people meeting any of the thrombotic thrombocytopenia purpura and hemolytic uremic syndrome screening criteria. The first date that a person met one of the screening criteria was referred to as the index date.
Patients with a plasmapheresis procedure code and without a thrombotic thrombocytopenia purpura or hemolytic uremic syndrome diagnosis code were removed if the physician service code in the 3 months preceding the plasmapheresis was for another indication (macroglobulinemia, cryoglobulinemia, multiple myeloma, Guillain-Barré syndrome, Lambert-Eaton syndrome, autoimmune hemolytic anemia, aplastic anemia, pure red cell aplasia, Goodpasture's syndrome, or myasthenia gravis).
A hematologist/oncologist (JK) on the study team reviewed each of the electronic data histories and determined whether the sequence of events was consistent with thrombotic thrombocytopenia purpura or hemolytic uremic syndrome. Patients who clearly did not have thrombotic thrombocytopenia purpura or hemolytic uremic syndrome cases based on electronic profiles were removed from the case group.
The total numbers of potential cases of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome were 343 for the United States, 56 for the GPRD, and 220 for Saskatchewan. We sought to review the medical records of all potential cases to confirm the diagnosis of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome.
Of the 619 medical records sought, we obtained 528 adequate abstractions (85%). For the remaining potential cases, a determination of case status was based on a medical care history as reflected in the chronologic claims listings or the computerized medical record considering the presence of diagnoses and procedures surrounding the index date that were consistent with thrombotic thrombocytopenia purpura or hemolytic uremic syndrome. Only incident cases were considered in the analysis.
Reasons for failure to obtain complete medical record abstractions included unavailability, inaccessible patient authorization signatures, and discrepancies between the computerized and written medical records.
Patient-identifying information was retained in secured files by each data holder and was physically separate from the research records.
Preonset Correlates of Disease
We examined the 3 months before the onset of disease, as defined by medical record review, for the presence of diagnoses, procedures, dispensings, or codes for the following known etiologies: cancer, bone marrow transplant, pregnancy, injection or dispensing of certain drugs (mitomycin C, cyclosporine A, FK-506, disease-modifying antirheumatic drugs), Escherichia coli infection, and HIV infection.
Incident cases were defined as the first occurrence of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome in the patient's record with no evidence of prior diagnosis in electronic or written records.
Each accepted case was classified by age, sex, and calendar year on index event. We calculated overall incidence rates, stratified by sex, specific time periods (1975–1989, 1990–1994, and 1995–2000 as possible in each dataset), and by known etiology. Incidence rates were calculated as the number of new cases of thrombotic thrombocytopenia purpura, hemolytic uremic syndrome, or both divided by the person-years of observation. We presented the age–sex standardized incidence rates using the age and sex distribution of the 1990 U.S. population.
Cause of Death
We obtained case vital status through December 31, 2000, in each database. For the United States, we identified causes of death by searching the National Death Index. For Saskatchewan, causes of death were derived from provincial vital statistics files, and for the GPRD, causes of death were obtained from the computerized medical records.
We chose at random up to 10 controls for each case matching the age, sex, and calendar year distribution of the cases. For each control, we assigned a random date during the year of the index event, and as of that date, we ascertained the same data items as for the cases of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome (diagnoses and medical procedures associated with all physician visits and hospitalizations, as well as all drugs dispensed for the prior 90 days). We divided the 90-day period into periods 1–21 and 22–90 days before the date of disease onset.
We examined every medical diagnosis or procedure present in at least 10% of the cases of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome, and any drug dispensed within 90 days before the date of disease onset in at least 5% of the cases. In addition, we investigated stent placements, percutaneous transluminal coronary angioplasty, dialysis, and angiography within 90 days before the index date.
We conducted a nested case-control study to compare the cases with the population sample on the same variables using a logistic regression procedure stratified by country to identify risk factors for thrombotic thrombocytopenia purpura or hemolytic uremic syndrome. We fitted separate logistic regression models to predict thrombotic thrombocytopenia purpura and hemolytic uremic syndrome.
Of the 528 charts reviewed, 280 were not a case, 207 were incident cases, 20 were recurrent cases, and 21 were patients with a history of thrombotic thrombocytopenia purpura with the documented event lying outside the study period. Five of the 91 possible cases in which only the computerized data were available were classified as incident cases.
Table 1 presents the demographic characteristics of the cases in the 3 populations. Women predominated in the thrombotic thrombocytopenia purpura and hemolytic uremic syndrome cases in the U.S. and Saskatchewan populations (64% and 73%, respectively). In the United Kingdom, 52% were women.
Corresponding to the period of population availability, most cases occurred between 1995 and 2000 for the United States, whereas for the United Kingdom and Saskatchewan, most cases occurred earlier.
Cancer was the most common known etiology for thrombotic thrombocytopenia purpura in the United States and Saskatchewan. E. coli infection was the most common known etiology for hemolytic uremic syndrome in all 3 regions.
As expected, most hemolytic uremic syndrome cases were age 19 years or younger. In the United States, 51% of the hemolytic uremic syndrome cases were younger than 10 years of age. In the United Kingdom, 73% of the cases were under the age of 10 years (64% under the age of 4). In Saskatchewan, 77% of the cases were below the age of 10 (68% were younger than 4; data not shown).
All 8 cases with multiple etiologies had cancer. Five cases (thrombotic thrombocytopenia purpura) had a preceding bone marrow transplant, 2 cases (one thrombotic thrombocytopenia purpura and one hemolytic uremic syndrome) had also taken a drug for cancer treatment, and one hemolytic uremic syndrome case both had a bone marrow transplant and had taken a drug for cancer treatment. In Saskatchewan, cancer (n = 2) and pregnancy (n = 1) were the only known etiologies other than E. coli infection preceding the diagnosis of thrombotic thrombocytopenia purpura or hemolytic uremic syndrome.
Twenty-one of the 60 thrombotic thrombocytopenia purpura cases were found to have died (35%), as had 5 of the 38 hemolytic uremic syndrome cases (13%). Of the 46 deceased cases in all 3 cohorts, thrombotic thrombocytopenia purpura or hemolytic uremic syndrome was recorded as the cause of death for 11, and in another 6 cases, thrombotic thrombocytopenia purpura or hemolytic uremic syndrome was a possible contributing cause of death.
Standardized Incidence Rates
Table 2 presents the standardized incidence rates of thrombotic thrombocytopenia purpura, hemolytic uremic syndrome, and both combined. The overall rates for thrombotic thrombocytopenia purpura were 3.8 per million per year in the United States and 1.2 per million per year in Saskatchewan. (There were only 2 thrombotic thrombocytopenia purpura cases in the U.K. data.) The overall rates for hemolytic uremic syndrome were 2.7 per million per year in the United States and 2.1 per million per year in the United Kingdom and in Saskatchewan. The overall incidences of thrombotic thrombocytopenia purpura (United States and Saskatchewan) and hemolytic uremic syndrome (United States, United Kingdom, and Saskatchewan) were higher for women.
The incidence of thrombotic thrombocytopenia purpura increased with age in the United States and peaked between 20 and 49 years in Saskatchewan. The incidence of hemolytic uremic syndrome for all 3 cohorts was highest in those under the age of 20 years. Except for hemolytic uremic syndrome cases in the United States, all incidence rates of thrombotic thrombocytopenia purpura and hemolytic uremic syndrome are greater between the years 1990 and 1994 in the 3 cohorts.
Predictors of Thrombotic Thrombocytopenia Purpura
Table 3 presents the predictors for thrombotic thrombocytopenia purpura found in the 21 days before the date of disease onset in all 3 datasets.
Diagnoses of hypertension (United States) and cancer (United States and Saskatchewan) were in excess among thrombotic thrombocytopenia purpura cases compared with the controls. All procedures occurring in over 10% of the cases in the United States were significant predictors of outcome. All drugs dispensed to 5% or more of the cases in the United States were significant predictors of disease. Dispensings of trimethoprim sulfamethoxazole, cardiac drugs, gastrointestinal drugs, antibiotics, conjugated estrogens, diazepam, and aspirin were strongly associated with thrombotic thrombocytopenia purpura (odds ratio ranging from 9 to infinity). No cases were preceded by percutaneous cardiac interventions, and none occurred in patients on dialysis.
Predictors of Hemolytic Uremic Syndrome
Table 4 presents the predictors for hemolytic uremic syndrome found in the 21 days before the date of disease onset in all 3 datasets.
Diagnoses of cancer (United States) and diarrhea (United Kingdom), which could be indicative of known causes of disease (ie, E. coli), were predictors of hemolytic uremic syndrome. The only procedure strongly associated with hemolytic uremic syndrome was bone marrow transplantation. Dispensings of ranitidine, furosemide, and amoxicillin were also strongly associated.
The incidence of thrombotic thrombocytopenia purpura that we found in the United States (3.8 per million person-years) is approximately 4 times higher than previously reported by Török10 for the years 1968 to 1991. In the U.K. population, the overall incidence of thrombotic thrombocytopenia purpura we found was considerably lower (0.1 per million person-years). In Saskatchewan, we found the overall incidence rate of thrombotic thrombocytopenia purpura was intermediate (1.2 per million per year), Overall, the incidence of thrombotic thrombocytopenia purpura was higher in women than in men.
The differences in thrombotic thrombocytopenia purpura incidence rates could indicate real differences in the populations we studied, but are more likely either the result of underrecording or the use of nonspecific diagnostic codes. In the GPRD, data are obtained from the general practitioner, but the diagnosis of thrombotic thrombocytopenia purpura will essentially always come from a consultant hematologist. It is possible that because this is a rare diagnosis made by specialists, it was not recorded or was coded nonspecifically in the clinician's files. The coding for outpatient files in the Saskatchewan dataset is restricted to the first 3 digits of an International Classification of Diseases, 9th Revision (ICD-9) code, making it difficult to identify potential cases treated in a nonhospital setting.
The standardized rates of hemolytic uremic syndrome were similar in the 3 datasets (2.7 per million person-years in the United States, 2.1 per million person-years in the United Kingdom, 2.1 per million per year in Saskatchewan). In all 3 regions, most cases were under the age of 20, and the disease occurred more often in females than in males. The incidence of hemolytic uremic syndrome that we found in the United States for those less than 20 years of age (males, 4.8 per million person-years; females, 6.7 per million person-years) was similar to the estimate of 6.7 per million children annually reported by Cummings et al. in California.11
Most of the cases of thrombotic thrombocytopenia purpura and hemolytic uremic syndrome we found in all 3 populations had no evidence of a previously reported etiology. Nevertheless, our case-control analysis provides further evidence that the known etiologies (cancer, pregnancy, and bone marrow transplant) are associated with thrombotic thrombocytopenia purpura and hemolytic uremic syndrome. Most of the relative risk estimates presented in our study have large confidence intervals, reflecting the low incidence of the diseases under study and the low frequency of exposure to the various risk factors in the general population (except pregnancy).
In the case-control analysis, diarrheal illness predominated as a risk factor for hemolytic uremic syndrome in the United Kingdom, and E. coli gastrointestinal infection was a common diagnosis found in the medical records of hemolytic uremic syndrome cases in all 3 regions. These results support previous findings that infection with E. coli O157:H7 is a cause of hemolytic uremic syndrome.1
We found only 2 ticlopidine users and one clopidogrel user in the 21 days preceding the thrombotic thrombocytopenia purpura diagnosis in the U.S. cohort, contributing minimally to the overall incidence of thrombotic thrombocytopenia purpura. Several of the drugs we found to be associated with thrombotic thrombocytopenia purpura or hemolytic uremic syndrome (estrogens, clarithromycin) have previously been reported as risk factors for these diseases, but the others have not.5 Additional evidence will be necessary to further corroborate the associations between these drugs and thrombotic thrombocytopenia purpura or hemolytic uremic syndrome.
Invasive procedures reported to be associated with thrombotic thrombocytopenia purpura (stents, percutaneous transluminal coronary angioplasty, and angiography) were not identified in any of the cases or controls. It is unlikely that these would have been missed, given the administrative and financial purposes of the source data in the United States and Saskatchewan, and given the centrality of general practitioner care in the United Kingdom. It appears that such procedures play only a small role in thrombotic thrombocytopenia purpura overall.
We appreciate the insightful advice of Mary Rose Stang from Saskatchewan Health and of David Goldsmith, who was also instrumental in initiating this research. We are grateful for the early contributions made to the study by Mei-Sheng Duh. This study is based in part on deidentified data provided by the Saskatchewan Department of Health. The interpretation and conclusions contained here do not necessarily represent those of the Government of Saskatchewan or the Saskatchewan Department of Health.