Study setting

This was a nationwide retrospective case-control study with a source population of approximately 9.4 million people between 2009 and 2010 [26]. It is built upon original methodology as part of the Vaccine Adverse Event Surveillance and Communication (VAESCO), European Centre of Disease Control [27].

Case and control identification

Cases were identified through diagnostic sleep centres and/or neurophysiology labs in Sweden, with the six largest centres (out of seven nationwide) participating. The centres provided a list of potential cases i.e. a list of individuals referred for a Multiple Sleep Latency Test (MSLT) during the study period, 1^st January 2009 and 31^st December 2010. Of the 431 potential cases, 142 received a primary diagnosis of narcolepsy by the initial referring clinician. Karolinska Institutet contacted these preliminarily diagnosed cases to request their participation in the study.

Of the 142 preliminary diagnosed cases, 27 refused participation and 18 were not contactable. The medical charts of the remaining 97 subjects were reviewed by a neurologist blinded to A(H1N1)pdm09 vaccination status. Consequently, 17 subjects were excluded as they were discovered to have an MSLT referral date outside the study period and eight did not have a diagnosis of narcolepsy according to the Brighton Classification Criteria [28]. This left, a total of 72 confirmed cases for inclusion in the study (Fig 1). The date when a case was first referred for a MSLT was defined as the index date.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart to show case and control identification.

https://doi.org/10.1371/journal.pone.0154296.g001

Controls were selected to match each case on age, gender, index date (so that the control did not have an MSLT-referral or narcolepsy diagnosis before its matched case), and county of residence. These were randomly selected from the Swedish Total Population Register, held by Statistics Sweden [29]. Four matched controls per case were invited to participate in the study. Of these, 251 individuals accepted the invitation (1057 did not respond, 152 refused participation, a further 156 were excluded, as they were not matched to cases that agreed to be part of the study, and four subsequently dropped out). (Fig 1).

Ethical approval

This study was approved by the regional ethics committee in Stockholm and written informed consent was collected from all study participants.

Data collection

Information regarding MSLT referral date and diagnosis date was collected from medical charts for the cases. Additional information for onset of cataplexy, excessive daytime sleepiness and Hypocretin-1 measurement in CSF was also available, however due to incomplete/missing/unknown dates for some cases these variables were excluded from the analysis.

Using a standardised questionnaire, a research nurse collected self-reported information, via telephone interviews, on previous disease and vaccination history from cases and controls for the period 1^st January 2005 - 31^st December 2010. Additional information on methodology can be found in the ECDC technical report [27]. Due to this long ascertainment period, information on self-reported medications and diseases was often incomplete i.e. the answer to simple yes/no questions regarding use of medication and diseases was missing in up to 45% of respondents, and the corresponding dates of use/disease were missing for between 73% and 100% (data not shown).

Consequently, self-reported information on diseases and medications was discarded in favour of register data, with the exception of vaccination status, which was only available through interview as the registration of vaccinations was not mandatory during the study period. Information on medication use was collected from the Prescribed Drug Register (PDR), which includes data on drugs prescribed in Sweden from July 2005 including prescription date, dosage and duration, expenditure and reimbursements. All drugs are classified according to the World Health Organisation Anatomical Therapeutic Chemical (ATC) Classification [30]. Disease information was collected from the National Patient Registers (NPR), which contains information on diagnoses in Swedish hospitals from specialised outpatient care from 2001 and inpatient care since 1987. The registers are managed by the National Board of Health and Welfare [31] and linkage of participant data to the registers is possible through their unique personal identification numbers [32].

Exposures

Data on the following ATC codes were collected to investigate whether prescription history was a risk factor for narcolepsy before A(H1N1)pdm09 vaccination: nervous system disorders (N), obstructive airway diseases (R03), antibacterial (J01), anti-inflammatory/anti-rheumatic drugs, non-steroid (M01A), antiviral (J05), immunosuppressant (L04) and diabetes (A10). Prescriptions for psychostimulants, agents used for ADHD treatment and nootropics (ATC N06BA01, N06BA03, N06BA04, N06BA07, and N06BA09) were used as a proxy for ADHD diagnosis in this study. The ADHD subcategory was created from prescriptions for N and therefore prescriptions for ADHD are included in the total number of prescriptions for N. Prescription issue date is used as a proxy for date of drug use.

Data on the following diagnosis codes, based on the International Classification of Disease (ICD) version 10, were collected from the NPR to investigate whether diagnosis history was a risk factor for narcolepsy before A(H1N1)pdm09 vaccination: nervous system disorders (G00-473, G475-99 –excluding narcolepsy (G474)), mental and behavioural disorders (F00-99), bacterial diseases (A30-49), neoplasms (C00-99), respiratory diseases (J00-99), viral infections of the CNS (A80-99), other infections of CNS (B25-34), other bacterial, viral and other infectious agents (B95-99) and diabetes mellitus (E10-14).

To investigate if a combination of prescription (ATC) and disease (ICD10) history was a risk factor for narcolepsy before A(H1N1)pdm09 vaccination, four combined exposure groups were created: a) nervous system disorders (ATC N and/or ICD10 G00-473, 475–99); b) bacterial diseases (ATC J01/ ICD10 A30-49); c) respiratory diseases (ATC R03/ICD10 J00-99) and; d) viral diseases (ATC J05/ ICD10 A80-99, B25-34).

For the purpose of measuring a crude measure of intensity a multiple prescription/diagnosis variable was created. This variable took into account that participants may have received one or more prescriptions and/or diagnoses over the exposure period in more than one prescription or diagnosis category.

Exposure windows

Disease history was studied as a risk factor for narcolepsy over three different exposure windows (Fig 2): 1) prescription (ATC) and diagnosis (ICD10) history before the index date. This period covered at least 5-years prior to the index date from the first prescription in the PDR (2005), or diagnosis date in the inpatient (1987) or outpatient (2001) registers respectively; 2) prescription history during the ‘vaccination period’ defined as six months before each case/control was first vaccinated with A(H1N1)pdm09 vaccine until one month after, with the exception of acute infections, e.g. prescriptions for antibacterials (ATC J01) and antivirals (ATC J05), where this period was shortened to two weeks before and after, as these infections are self-limiting; 3) prescription and diagnosis history before A(H1N1)pdm09 vaccination—again covering at least 5-years prior to vaccination, from first prescription or diagnosis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Exposure windows.

1. Prescription (ATC) and diagnosis (ICD10) history before index date (MSLT-referral date). Ever was defined as after the first date in either the inpatient register (1987), outpatient register (2001) or prescribed drug register (2005) and prior to the index date. 2. Prescription (ATC) and diagnosis (ICD10) history during the vaccination period. This period extended to six months before to one month after A(H1N1)pdm09 vaccination (specific to each individual). For acute infections (bacterial (ATC J01) and viral (ATC J05)) this period was shortened to two weeks before to two weeks after vaccination. 3. Prescription (ATC) and diagnosis (ICD10) history before vaccination. Ever was defined as after the first date in either the PDR (2005), inpatient register (1987) or outpatient register (2001) and prior to the vaccination date.

https://doi.org/10.1371/journal.pone.0154296.g002

The self-reported A(H1N1)pdm09 vaccination dates were incomplete for some cases and controls. In such instances the day of month was assigned the 15^th, missing months in 2009 were assigned to November i.e. the middle of October-December, and missing months in 2010 were assigned to January i.e. the middle of the overall vaccination period of October 2009-March 2010.

Statistical analysis

Conditional logistic regression was used to estimate odds ratios (OR) together with 95% confidence intervals (95% CI). Given the rarity of narcolepsy in the general population, ORs were interpreted as the relative risk (RR) of narcolepsy [33]. Each exposure was assessed as a covariate in a univariate model, which was controlled for the joint effect of the matching variables through conditioning. The significance level was 0.05 and all tests were two sided. The analyses were performed using SAS version 9.4.

Three analyses were performed. First, a full data analysis investigating disease history as a risk factor for narcolepsy. In this model, all cases and controls were included regardless of vaccination status and timing of vaccination, i.e. whether the index date was before A(H1N1)pdm09 vaccination or after. The exposure window of interest in this analysis was prescriptions or diagnoses before the index date (exposure window 1, Fig 2).

Second, a vaccinated-only analysis i.e. cases and controls vaccinated before the index date, to determine if disease history is an effect modifier for vaccination status. The periods of interest were prescriptions and diagnosis history before the index date (exposure window 1), prescription history during the vaccination period (exposure window 2) and prescription and diagnosis history before vaccination (exposure window 3, Fig 2).

Third, a case-only analysis was used to compare vaccinated cases with an index date before vaccination to vaccinated cases with an index date after vaccination. This analysis is based on the assumption that for a rare disease, with somewhat unspecific symptoms, there may be an extended history of misdiagnoses/treatment prior to the correct diagnosis, leading to confounding by indication. The exposure windows of interest in this analysis included prescription and diagnosis history during the vaccination period and before vaccination (exposure window 2 and 3 respectively, Fig 2).

Sensitivity analyses using the diagnosis date of narcolepsy as the index date instead of MSLT-referral date were carried out for all three analyses.

Strengths and limitations

To reduce the effects of confounding, strict definitions were used for inclusion criteria and we matched cases on age, gender, index date and county of residence. At the time of case ascertainment, Type-1 and Type-2 narcolepsy definitions had not been introduced. Due to incomplete/missing dates for symptom onset e.g. cataplexy, EDS and hypocretin-1 measurement for some cases, old case definitions were kept. This is a limitation of the study. However, it is unlikely that the lack of this information would have changed the results, nor would it have helped us answer whether previous disease history was a risk factor for narcolepsy. To compensate for missing dates and reduce recall bias in the self-reported information, we used exposure information from the Swedish healthcare registers. The A(H1N1)pdm09 vaccination date was self-reported however. Although we controlled for self-reported vaccination dates we cannot rule out the possibility of recall bias, with cases being more likely to remember if and when they received their vaccination than controls, resulting in an overestimation of exposure in cases. However, as this study took place shortly after the vaccination campaign we believe this reduced the potential effects of recall bias. To account for the time between vaccination and diagnosis, several exposure windows were analysed specific to each individual and based on/around the date they first received the A(H1N1)pdm09 vaccine, or their index date.

Diagnostic bias was a potential issue in this study as typically, many narcolepsy cases have an incremental progression of symptoms over several years, which results in a lag between the onset of symptoms and diagnosis. However, following the vaccination campaign there was an increased awareness of narcolepsy, which shortened the time between the onset of symptoms and diagnosis. To reduce the potential for diagnostic bias in this study, the MSLT-referral date was used as the index date and not the date of diagnosis.

It is of note that the ATC and ICD codes used in this study are only proxies for a number of diseases and they have broad underlying aetiologies, so while they can provide estimations for disease history they are not as specific as individual codes. In addition, people are less likely to seek medical attention for milder infections, which could have resulted in an underestimation of such exposures in this study. It was unfortunately not possible to look at individual codes in this study, due to a limited sample size. While lack of power is a limitation in this study, it still remains one of the larger observational studies to date in Sweden, which looks at the association between A(H1N1)pdm09 vaccination and narcolepsy.