Risk of neuropsychiatric adverse events associated with the use of oseltamivir: a nationwide population-based case-crossover study

Abstract

Background

Although the potential risk of neuropsychiatric adverse events (NPAEs) upon administration of oseltamivir has been raised in case reports, the association between the use of oseltamivir and the risk of NPAEs is unclear.

Objectives

We aimed to evaluate whether the use of oseltamivir triggers NPAEs.

Patients and methods

We conducted a population-based case-crossover study using the National Sample Cohort data from the National Health Insurance Service in South Korea. From a total of 236 348 incident patients with NPAEs as either a primary or secondary diagnosis, our final case series included 5322 patients with a prior prescription for oseltamivir between 2009 and 2013. Exposure to oseltamivir was assessed during 2, 7, 14, 28 and 56 day hazard periods prior to each patient’s NPAE. Three pre-consecutive control periods were matched using the same time windows. Conditional logistic regression analysis was used to estimate adjusted ORs (aORs), adjusting for time-variant diagnosis of influenza and concomitant medications.

Results

Matched analyses found a consistently increased risk of NPAEs associated with the use of oseltamivir in the 2 day (aOR 1.90, 95% CI 1.29–2.81), 7 day (aOR 1.32, 95% CI 1.00–1.74), 14 day (aOR 1.28, 95% CI 1.03–1.60), 28 day (aOR 1.25, 95% CI 1.06–1.47) and 56 day (aOR 1.13, 95% CI 0.99–1.29) hazard periods compared with use in the three control periods.

Conclusions

This study found that the short-term use of oseltamivir triggers the incidence of NPAEs. Early monitoring of NPAEs may be required when prescribing oseltamivir with careful consideration of the risk–benefit balance of oseltamivir.

Introduction

Antiviral medications have been shown to be effective for the prevention and treatment of influenza by reducing the duration and severity of illness.¹ Oseltamivir is a commonly prescribed antiviral drug for the treatment of influenza. However, concerns have been raised regarding potential neuropsychiatric adverse events (NPAEs) following oseltamivir use. Case reports from the USA, Spain, Japan, China and South Korea have suggested that NPAEs, including delirium, depressive episodes, mania, confusion syndrome, suicidal feelings and hallucinations, may occur after oseltamivir administration.^2–7 Although the US FDA updated the oseltamivir package insert to include the abrupt onset and sometimes fatal nature of NPAEs in 2008,⁸ the US FDA’s Adverse Event Reporting System (FAERS) data between 1999 and 2012 suggest a disproportionate increase in the reporting of NPAEs, with reporting ORs (RORs) of 29.35 (95% CI 27.43–31.40) for abnormal behaviour and 15.36 (95% CI 14.39–16.39) for psychiatric and behavioural symptoms.⁹ FAERS post-marketing data in 2015 repeatedly reported an ROR of 3.8 (95% CI 3.5–4.0) for NPAEs compared with all other drugs, which increased in patients younger than 20 years, suggesting that oseltamivir may be associated with NPAEs.¹⁰

Despite safety concerns regarding oseltamivir use, a meta-analysis that integrated placebo-controlled randomized trials reported no significant association between oseltamivir use and neurological or psychiatric disorders.¹¹ However, the risk of bias in industry-funded oseltamivir trials was raised in a previous meta-analysis that included trials sponsored by a specific company, potentially affecting the reliability of the results.¹² A subsequent systematic review of placebo-controlled randomized trials also found a non-significant association between oseltamivir use and the risk of psychiatric adverse events in treatment trials, reporting a risk ratio of 0.93 (95% CI 0.43–2.03), compared with a risk ratio of 1.80 (95% CI 1.05–3.08) in prophylaxis trials.¹³ It is possible that NPAEs were a rare occurrence in the population included in the randomized trials and the power was insufficient to detect an association. The confidence interval was wide (95% CI 0.43–2.03) and did not rule out a doubling in risk as a result of treatment, as was found in the prophylaxis trials.

In 2009, a cohort study including ∼10 000 children reported that abnormal behaviour was associated with oseltamivir use, showing a rate ratio of 1.57 (95% CI 1.34–1.83) in Japan.¹⁴ However, another cohort study published by the US CDC after the 2009 influenza A (H1N1) pandemic reported no evidence of an increased risk of NPAEs following oseltamivir treatment, with ORs of 1.21 (95% CI 0.74–1.97) and 0.94 (95% CI 0.66–1.34) for incident psychiatric events during the 1–7 day and 1–14 day risk intervals, respectively.¹⁵ Despite a cohort of over 27 000 matched pairs, the number of observed adverse events was low and confidence intervals were wide due to the rare outcome. A further limitation of the study was that no prescription medications other than influenza antiviral medications were considered as factors potentially influencing the outcome.

The aim of this study was to explore whether oseltamivir use is associated with subsequent NPAEs using a case-crossover study design, which is a method used to eliminate between-person time-invariant confounders.

Patients and methods

Ethics

This study was approved by the Institutional Review Board of the Sungkyunkwan University in South Korea (SKKU-IRB-2017-07-012). All personal identifying information for patients was removed; therefore, informed consent was waived by the Institutional Review Board for this study.

Database

This study used the National Sample Cohort data from the National Health Insurance Service (NHIS-NSC) in South Korea. The NHIS uses a systematic sampling approach to randomly select a representative database of ∼1 million people between 2002 and 2013, which is 2.2% of the total population. The sample cohort was compared with the entire population according to the average total annual medical expenses, residence distribution and the mean and standard deviation of health insurance premiums, and the differences were negligible during cohort years.¹⁶ The data give researchers access to demographic data, including sex, age recorded at 5 year intervals, income level and date of death, as well as healthcare data, including clinical diagnoses, medical procedures, expenditures and drug prescriptions. Information on prescribed drugs included the generic drug name, prescription date, duration and route of administration.

Case-crossover design

A case-crossover design was used to assess the association between oseltamivir exposure and the risk of NPAEs, using the study subjects at previous timepoints as their own controls.¹⁷ The case-crossover design has been widely used as a tool to evaluate drug safety and is particularly suitable when the exposure is intermittent, when the effect on risk is transient and when the outcome is abrupt.¹⁸ In this design, only patients experiencing an event of interest were included and their exposures were measured during the hazard and control periods. Accordingly, the number of patients exposed to oseltamivir in the hazard period (which is the period immediately before the event of interest) is compared with the number of those exposed to oseltamivir in the control period (which is a period prior to but of the same length as the hazard period).¹⁹ Instead of matched controls, using study subjects as their own controls enabled us to reduce confounding by subject characteristics that did not change over time, including characteristics that were difficult to investigate or impossible to measure.²⁰

A positive association between the NPAEs and current use of the oseltamivir may be found in patients who had not used oseltamivir in the control period but had used it in the hazard period. A negative association between the NPAEs and oseltamivir use would be expected with oseltamivir use in the control period but not during the hazard period. Case-crossover studies only deal with such discordant pairs between hazard and control period. Concordant pairs, in which the patient is administered oseltamivir in both the hazard and control period, or does not use oseltamivir in either period, do not contribute to the estimation of the OR in case-crossover studies.

Case definition

We identified all patients who had been prescribed oseltamivir on at least one occasion prior to the NPAE diagnosis (primary or secondary) between 1 January 2009 and 31 December 2013, taking into account that the use of oseltamivir increased after the 2009 influenza A (H1N1) pandemic.²¹

Based on the NPAE categories reported in previous studies,^13,22 a psychiatric clinician (W. J. Kim) selected ICD-10 codes appropriate for each category (Table S1, available as Supplementary data at JAC Online). Patients with corresponding ICD-10 codes were defined as cases when the codes were recorded as a primary or secondary diagnosis during hospitalization or outpatient visits. Our cases were defined as subjects with any NPAE including confusion, hallucinations, paranoia, psychosis, schizophrenia, anxiety, restlessness, nervousness, depression, bipolar disorder, stress symptoms, sleeping disorder, aggression, suicide ideation, convulsion, encephalitis, ataxia, dizziness and giddiness, and vertigo. The index date was defined as the date of the first NPAE diagnosis between 1 January 2009 and 31 December 2013.

The following exclusion criteria were applied: (i) pre-existing NPAEs in 2008 to confirm incident cases of NPAEs; (ii) diagnosis of chronic kidney disease (ICD-10 codes N18.3, N18.4, N18.5, N18.6 and N19) in 2008, as this condition can affect the rate of drug clearance, complicating hazard and control period definitions; and (iii) diagnosis of NPAE on the same day as the first oseltamivir was prescribed as the time association is not certain in the claims database (Figure 1).

Figure 1.

Study subject selection.

Exposure to oseltamivir

Exposure to oseltamivir was defined as a patient receiving at least one prescription for the medication (Anatomic Therapeutic Chemical code J05AH02). The prescriptions from both inpatient and outpatient settings were assessed and the exposures were measured during the hazard and control periods. We set a 14 day hazard period in our main analyses, as a cohort study reported a median time from influenza diagnosis to the first CNS outcome of 13 days.²³ Moreover, another cohort study applied a 14 day period to ensure that acute events were captured.²⁴

We also applied different hazard period time frames to examine whether the results changed significantly according to the different time windows. The 7 day risk interval was used to reflect the 5 day treatment course and the half-life of the active metabolite (6–10 h) in patients with normal renal function.¹⁵ Many studies reported an abrupt onset and rapid resolution of NPAEs, with onset frequently occurring after a single oseltamivir dose; therefore, the risk interval could also be shortened to 2 days following the drug prescription.¹⁵ We also analysed the 28 and 56 day risk intervals to examine how the risk changes with a longer follow-up duration without missing any further events.

We set three consecutive control periods of the same duration as the hazard period for each case. To prevent any carry-over effect, a washout period of the same duration as each hazard period was chosen between the control and the hazard period (Figure 2).

Figure 2.

Design of case-crossover study of oseltamivir and NPAEs. The hazard period was defined as the 2, 7, 14, 28 and 56 day time window before the first diagnosis of an NPAE. The same washout period duration was chosen between the hazard and the control period. Three consecutive control periods included 2, 7, 14, 28 and 56 day time windows.

To determine that our model appropriately controlled for seasonality, we performed a sensitivity analysis with three additional control periods and a 14 day hazard period. The first control period used in each sensitivity analysis started from 11, 12 and 13 months before the index date, respectively, and included the preceding 14 days. Each first control period was followed by two consecutive 14 day control periods.

Time-varying confounders

The study design controlled for time-invariant confounders, such as genetic factors and sex. However, there are also confounders which may change between hazard and control period. For each patient, concomitant medications that could potentially change across the hazard and control periods were considered as within-patient time-varying confounders and included in the adjusted analysis. Medications frequently associated with neuropsychiatric disorders included non-steroidal anti-inflammatory drugs (NSAIDs), opioid drugs, benzodiazepines, antidepressants, methylphenidate, zolpidem, antiepileptics, histamine-2 receptor antagonists, calcium channel blockers, corticosteroids and statins (Table S2).^25–27 Among these, significant differences in the distribution of medications prescribed during the hazard and control periods were observed for NSAIDs, opioid drugs, benzodiazepines, antidepressants, zolpidem and corticosteroids (Table S3). Considering that the cases served as their own controls in the case-crossover design, we included the concomitant medications showing different distributions in the hazard and control periods as time-varying confounders. Furthermore, to adjust for the effects of influenza virus infection on the NPAEs,^28,29 the diagnosis of influenza (ICD-10 codes J09, J10 and J11) was included as a time-varying confounder.

Statistical analysis

Descriptive statistics were used to present the characteristics of the study subjects. The distribution of high-risk conditions for severe influenza complications within 1 year from the index date was assessed by the ICD-10 codes (Table S4).¹ Comorbidities and concomitant medications recorded on at least one occasion within 1 year of the index date were included in the analysis. An influenza season for each patient was defined based on the date of prescription of oseltamivir, from July of the year in which it was prescribed to June of the following year.

We conducted conditional logistic regression analysis to estimate crude and adjusted ORs (aORs), by comparing the odds of oseltamivir exposure between the hazard period and three consecutive control periods before the NPAEs. Our final model included exposure status, time-variant record of influenza, concomitant use of NSAIDs, opioids, benzodiazepines, antidepressants, zolpidem, histamine-2 receptor antagonists and corticosteroids. Methylphenidate, antiepileptics, calcium channel blockers and statins were not included in the final model due to the lack of significant difference in distribution between the hazard and control periods. To rule out the possibility that the excluded medications would affect the goodness of fit of the model, we calculated the difference in likelihood ratio between the final model and a new model that incorporated one of the excluded medications as an additional covariate in the final model. For each of the four excluded medications, the difference in the likelihood ratio was not statistically significant.

A test of interaction was performed to examine whether the difference in effect size was statistically significant depending on age group, sex, Charlson comorbidity index and presence or absence of high-risk conditions for severe influenza complications.³⁰ All tests were two-sided, with a significance level of 0.05. All data transformations and statistical analyses were conducted using SAS version 9.4 for Windows (SAS Institute, Cary, NC, USA).

Results

We identified 5322 individuals diagnosed with an NPAE with oseltamivir use prior to the NPAE between 1 January 2009 and 31 December 2013 (Figure 1). Overall, 60.54% (3222/5322) of subjects were female and the age distribution was as follows: 53.63% aged 20–64 years, 20.61% aged 10–19 years and 15.25% aged 0–9 years. The following high-risk medical conditions that may be associated with the incidence of severe influenza complications were frequently observed: asthma (22.60%), metabolic disease (21.33%) and liver disease (16.20%). There were no differences in the distribution between male and female subjects for the majority of pre-defined high-risk medical conditions, although asthma and COPD were more prevalent in male subjects (P < 0.0001 and P = 0.0271, respectively). The probability of oseltamivir treatment was highest during the pandemic period, between July 2009 and June 2010, during which 87.09% (4635/5322) of subjects were prescribed oseltamivir (Table 1).

Table 1.

Characteristics of the study patients with oseltamivir use preceding NPAEs

Variable	No. (%)			Pa
	all patients (n = 5322)	male (n = 2100)	female (n = 3222)
Age group (years)
0–9	811 (15.3)	459 (21.9)	352 (10.9)	<0.0001
10–19	1097 (20.6)	527 (25.1)	570 (17.7)
20–64	2854 (53.6)	912 (43.4)	1942 (60.3)
≥65	560 (10.5)	202 (9.6)	358 (11.1)
High-risk condition for severe influenza complications within 1 year of index dateb
asthma	1186 (22.3)	538 (25.6)	648 (20.1)	<0.0001
COPD	87 (1.6)	44 (2.1)	43 (1.3)	0.0271
cardiovascular disease	273 (5.1)	120 (5.7)	153 (4.8)	0.0918
cardiac dysrhythmia	67 (1.3)	27 (1.3)	40 (1.2)	0.8407
diabetes	477 (9.0)	199 (9.5)	278 (8.6)	0.2203
metabolic diseases (excluding diabetes)	1119 (21.0)	410 (19.5)	709 (22.0)	0.0565
renal disease	271 (5.1)	103 (4.9)	168 (5.2)	0.7030
liver disease	850 (16.0)	333 (15.9)	517 (16.1)	0.9644
malignancy	202 (3.8)	74 (3.5)	128 (4.0)	0.4624
immunosuppressive disorder	4 (0.1)	2 (0.1)	2 (0.1)	0.6467c
neurological/musculoskeletal disease	300 (5.6)	120 (5.7)	180 (5.6)	0.7444
morbid obesity	1 (0.02)	0 (0)	1 (0.03)	>0.9999c
Charlson comorbidity index
0	2304 (43.3)	917 (43.7)	1387 (43.1)	0.0110
1	1738 (32.7)	721 (34.3)	1017 (31.6)
>1	1280 (24.1)	462 (22.0)	818 (25.4)
Concomitant medication within 1 year of index date
NSAIDs	4613 (86.7)	1768 (84.2)	2845 (88.3)	<0.0001
opioid drugs	2163 (40.6)	745 (35.5)	1418 (44.0)	<0.0001
benzodiazepines	1383 (26.0)	415 (19.8)	968 (30.0)	<0.0001
antidepressants	281 (5.3)	99 (4.7)	182 (5.6)	0.1363
methylphenidate	50 (0.9)	38 (1.8)	12 (0.4)	<0.0001
zolpidem	184 (3.5)	62 (3.0)	122 (3.8)	0.1036
antiepileptics	121 (2.3)	50 (2.4)	71 (2.2)	0.6714
histamine-2 receptor antagonists	3039 (57.1)	1013 (48.2)	2026 (62.9)	<0.0001
calcium channel blockers	100 (1.9)	45 (2.1)	55 (1.7)	0.2524
corticosteroids	3147 (59.1)	1192 (56.8)	1955 (60.7)	0.0045
statins	2124 (39.9)	793 (37.8)	1331 (41.3)	0.0098
Timing of influenza
January 2008–June 2009	67 (1.3)	35 (1.7)	32 (1.0)	0.0016
July 2009–June 2010	4635 (87.1)	1790 (85.2)	2845 (88.3)
July 2010–June 2011	199 (3.7)	76 (3.6)	123 (3.8)
July 2011–June 2012	353 (6.6)	164 (7.8)	189 (5.9)
July 2012–December 2013	68 (1.3)	35 (1.7)	33 (1.0)

a

The P value was calculated using the χ² test for categorical variables.

b

The number of patients diagnosed with haemoglobinopathy was omitted because it was zero.

c

The P value was calculated using Fisher’s exact test for categorical variables.

Figure 3 indicates the number of patients prescribed oseltamivir and the number of patients newly diagnosed with NPAEs between 2009 and 2013. The incidence of NPAEs remained constant regardless of the year and season, whereas the prescriptions of oseltamivir were mainly concentrated from winter to early spring each year.

Figure 3.

Number of patients newly diagnosed with NPAEs and number of patients prescribed oseltamivir between 2009 and 2013 for the 5322 cases included in this study. The solid trendline indicates the patients with incident NPAEs and the bar graph indicates the patients prescribed oseltamivir.

The use of oseltamivir within the 14 day time window was associated with a 1.28-fold higher risk of NPAEs after adjustment for the time-varying record of influenza and concomitant medications (95% CI 1.03–1.60; P = 0.0266). The risk remained elevated and was 1.32-fold higher in the 7 day (95% CI 1.00–1.74; P = 0.0468) and 1.90-fold higher in the 2 day time window (95% CI 1.29–2.81; P = 0.0012). However, the aOR decreased to 1.25 (95% CI 1.06–1.47; P = 0.0087) in the 28 day time window and became insignificant in the 56 day time window, with an aOR of 1.13 (95% CI 0.99–1.29; P = 0.0608) (Table 2).

Table 2.

Risk of NPAEs associated with the use of oseltamivir with different time windows (2, 7, 14, 28 and 56 days)

Time window (days)	No. (%)		Crude OR (95% CI)	P	aORa (95% CI)	P
	exposed in hazard period (n = 5322)	exposed in control period (n = 15 966)
2	61 (1.1)	101 (0.6)	2.26 (1.53–3.32)	<0.0001	1.90 (1.29–2.81)	0.0012
7	94 (1.8)	177 (1.1)	1.68 (1.29–2.20)	0.0001	1.32 (1.00–1.74)	0.0468
14	139 (2.6)	263 (1.6)	1.64 (1.32–2.03)	<0.0001	1.28 (1.03–1.60)	0.0266
28	230 (4.3)	480 (3.0)	1.46 (1.24–1.71)	<0.0001	1.25 (1.06–1.47)	0.0087
56	361 (6.8)	855 (5.4)	1.27 (1.12–1.44)	0.0001	1.13 (0.99–1.29)	0.0608

a

Conditional logistic regression adjusted for important potential time-varying confounding variables including diagnosis of influenza, prescription of NSAIDs, opioid drugs, benzodiazepines, antidepressants, zolpidem, histamine-2 receptor antagonists and corticosteroids.

The highest increased risk was observed among patients aged 10–19 years, with aOR of 2.27 (95% CI 1.22–4.22), followed by patients aged 0–9 years (aOR 1.25; 95% CI 0.78–1.99), patients ≥65 years (aOR 1.38; 95% CI 0.77–2.47) and patients aged 20–64 years (aOR 1.13; 95% CI 0.82–1.54), although the difference in risk by age group was not significant (P value for interaction = 0.2608). Male subjects had an aOR of 1.58 (95% CI 1.10–2.27) and the difference in risk according to sex was not significant (P value for interaction = 0.1692) (Table 3). None of the high-risk conditions for severe influenza complications resulted in significant effect modification (P values for interaction >0.05) (Table 4).

Table 3.

Risk of NPAEs associated with the use of oseltamivir over the preceding 14 days according to age group, sex and Charlson comorbidity index

Variable	n/N (%)		Crude OR (95% CI)	aORa (95% CI)	P value for interactionb
	exposed in hazard period	exposed in control period
Age group (years)
0–9	31/811 (3.8)	64/2433 (2.6)	1.52 (0.96–2.39)	1.25 (0.78–1.99)	0.2608
10–19	23/1097 (2.1)	24/3291 (0.7)	3.10 (1.71–5.63)	2.27 (1.22–4.22)
20–64	66/2854 (2.3)	136/8562 (1.6)	1.49 (1.10–2.01)	1.13 (0.82–1.54)
≥65	19/560 (3.4)	39/1680 (2.3)	1.50 (0.85–2.65)	1.38 (0.77–2.47)
Sex
male	55/2100 (2.6)	87/6300 (1.4)	2.01 (1.41–2.86)	1.58 (1.10–2.27)	0.1692
female	84/3222 (2.6)	176/9666 (1.8)	1.47 (1.12–1.92)	1.15 (0.87–1.52)
Charlson comorbidity index
0	46/2304 (2.0)	79/6912 (1.1)	1.18 (1.24–2.64)	1.36 (0.92–2.01)	0.0838
1	63/1738 (3.6)	98/5214 (1.9)	2.03 (1.46–2.83)	1.51 (1.07–2.13)
>1	30/1280 (2.3)	86/3840 (2.2)	1.05 (0.68–1.61)	0.91 (0.59–1.41)

a

Conditional logistic regression adjusted for important potential time-varying confounding variables including diagnosis of influenza, prescription of NSAIDs, opioid drugs, benzodiazepines, antidepressants, zolpidem, histamine-2 receptor antagonists and corticosteroids.

b

Likelihood ratio test for interaction, indicating whether oseltamivir affects NPAEs differently depending on age groups, sex and Charlson comorbidity index.

Table 4.

Risk of NPAEs associated with the use of oseltamivir over the preceding 14 days according to the presence or absence of high-risk conditions for severe influenza complications within 1 year of the index date

High-risk conditions for severe influenza complications		No. (%)		Crude OR (95% CI)	aORa (95% CI)	P value for interactionb
		exposed in hazard period	exposed in control period
Asthma	yes	36/1186 (3.0)	82/3558 (2.3)	1.35 (0.90–2.03)	1.13 (0.74–1.72)	0.4272
	no	103/4136 (2.5)	181/12408 (1.5)	1.77 (1.38–2.27)	1.36 (1.05–1.76)
COPD	yes	3/87 (3.4)	6/261 (2.3)	1.63 (0.35–7.63)	1.95 (0.40–9.55)	0.9046
	no	136/5235 (2.6)	257/15705 (1.6)	1.64 (1.32–2.03)	1.27 (1.02–1.59)
Cardiovascular disease	yes	7/273 (2.6)	13/819 (1.6)	1.62 (0.65–4.05)	1.32 (0.52–3.39)	0.8824
	no	132/5049 (2.6)	250/15147 (1.6)	1.64 (1.32–2.04)	1.27 (1.02–1.60)
Cardiac dysrhythmia	yes	2/67 (3.0)	6/201 (3.0)	1.00 (0.20–4.96)	1.11 (0.19–6.54)	0.6092
	no	137/5255 (2.6)	257/15765 (1.6)	1.66 (1.33–2.05)	1.28 (1.03–1.60)
Diabetes	yes	9/477 (1.9)	34/1431 (2.4)	0.79 (0.37–1.66)	0.77 (0.36–1.65)	0.0729
	no	130/4845 (2.7)	229/14535 (1.6)	1.77 (1.42–2.22)	1.34 (1.06–1.69)
Metabolic diseases, excluding diabetes	yes	27/1119 (2.4)	74/3357 (2.2)	1.10 (0.70–1.74)	0.92 (0.58–1.47)	0.0875
	no	112/4203 (2.7)	189/12609 (1.5)	1.86 (1.45–2.37)	1.42 (1.10–1.82)
Renal disease	yes	7/271 (2.6)	13/813 (1.6)	1.65 (0.64–4.21)	1.28 (0.48–3.39)	0.9811
	no	132/5051 (2.6)	250/15153 (1.7)	1.64 (1.32–2.04)	1.29 (1.03–1.61)
Liver disease	yes	28/850 (3.3)	53/2550 (2.1)	1.63 (1.02–2.62)	1.21 (0.74–1.97)	0.9710
	no	111/4472 (2.5)	210/13416 (1.6)	1.64 (1.29–2.09)	1.29 (1.01–1.66)
Malignancy	yes	4/202 (2.0)	13/606 (2.2)	0.92 (0.29–2.94)	0.89 (0.27–2.95)	0.3571
	no	135/5120 (2.6)	250/15360 (1.6)	1.68 (1.35–2.08)	1.32 (1.05–1.65)
Immunosuppressive disorder	yes	0/4 (0)	1/12 (8.3)	NA	NA	0.9408
	no	139/5318 (2.6)	262/15954 (1.6)	1.65 (1.33–2.04)	1.29 (1.03–1.60)
Neurological/musculoskeletal disease	yes	7/300 (2.3)	23/900 (2.6)	0.91 (0.38–2.16)	0.93 (0.39–2.27)	0.2385
	no	132/5022 (2.6)	240/15066 (1.6)	1.71 (1.37–2.14)	1.32 (1.05–1.65)

NA, not applicable.

a

Conditional logistic regression adjusted for important potential time-varying confounding variables including diagnosis of influenza, prescription of NSAIDs, opioid drugs, benzodiazepines, antidepressants, zolpidem, histamine-2 receptor antagonists and corticosteroids.

b

Likelihood ratio test for interaction, indicating whether there are differences in the effects of oseltamivir on NPAEs between patients with high-risk conditions and those without high-risk conditions after adjusting for concomitant medication.

Sensitivity analysis using 14 day control periods 11, 12 and 13 months before the index date yielded results consistent with the main results (Table 5). The number of patients for each NPAE is described in Table S5.

Table 5.

Sensitivity analysis of risk of NPAEs associated with the use of oseltamivir over the preceding 14 days using control periods (11, 12 and 13 months before the index date)

First control period	No. (%)		Crude OR (95% CI)	P	aORa (95% CI)	P
	exposed in hazard period (n = 5322)	exposed in control period (n = 15 966)
11 months before the index date	139 (2.6)	215 (1.3)	2.03 (1.63–2.54)	<0.0001	1.59 (1.27–2.00)	<0.0001
12 months before the index date	139 (2.6)	278 (1.7)	1.57 (1.27–1.95)	<0.0001	1.33 (1.07–1.66)	0.0118
13 months before the index date	139 (2.6)	196 (1.2)	2.24 (1.79–2.81)	<0.0001	1.73 (1.37–2.18)	<0.0001

a

Conditional logistic regression adjusted for important potential time-varying confounding variables including diagnosis of influenza, prescription of NSAIDs, opioid drugs, benzodiazepines, antidepressants, zolpidem, histamine-2 receptor antagonists and corticosteroids.

Discussion

Principal findings

Our population-based case-crossover study found a 1.28-fold increased risk of NPAEs within 14 days of oseltamivir exposure compared with other control periods. We altered the duration of the hazard time window to 2, 7 and 28 day periods and confirmed that the association between oseltamivir use and NPAEs was significant across varying time windows. In addition, the association remained significant when controlled for seasonality. The association was not affected by age groups, sex, Charlson comorbidity index and presence or absence of high-risk conditions for severe influenza complications, suggesting that the risk of NPAEs with oseltamivir use is significant irrespective of age group, sex, Charlson comorbidity index and presence or absence of high-risk conditions for severe influenza complications.

Comparison with other studies

Our findings were consistent with a previous cohort study showing a 1.21-fold increase in the risk of psychiatric events associated with the use of oseltamivir and a 1.5-fold increase in the risk for patients aged 10–19 years compared with oseltamivir-untreated patients.¹⁵ Although the 95% CI of each estimate was not statistically significant due to the nature of rare events in the cohort design, the 95% CIs for the estimates overlapped with our findings. A cohort study in children showed a 1.57-fold increase in the risk of abnormal behaviour associated with oseltamivir use compared with the no-use group and the 95% CI (1.31–1.83) for the estimate overlapped with our findings for children.¹⁴ Our findings are also consistent with those reported from the FAERS database, showing an increased adjusted ROR of 2.3 (95% CI 2.0–2.6) for NPAEs in male subjects, which substantially increased to 13.4 (95% CI 11.0–16.3) for male patients aged 10–19 years.¹⁰ The FAERS database consisted of post-marketing data which could not ascertain causal relationships; therefore, our results better reflect the clinical reality as causality was revealed using real-world data. A recent self-controlled case-series study found that abnormal behaviour was 1.9 to 29.1 times more likely with oseltamivir use. The increased risk was considerably higher than that reported in this study and may be explained by the small study population of 28 subjects and by the fact that the design could not exclude the possibility of abnormal behaviour caused by influenza itself.³¹

Previous cohort studies suggested a protective effect of oseltamivir, showing a 0.82-fold decreased risk of neuropsychiatric outcome³² and a 0.72-fold decreased risk of neuropsychiatric outcome.³³ However, the results may be biased because confounders such as comorbidities or concomitant medication, which could increase the risk of outcome, were not adequately adjusted for.

Biological mechanism

The possible mechanism underlying NPAEs in patients treated with oseltamivir was previously attributed to the influenza infection. However, our findings have shown significant results even after adjusting for the diagnosis of influenza and an increase in the penetration of oseltamivir from the plasma to the CNS may be a more plausible mechanism of action.

As oseltamivir phosphate is a prodrug, it dissociates in the gastrointestinal tract to form oseltamivir as a free base (OT), which is absorbed and metabolized into the active form of the neuraminidase inhibitor, oseltamivir carboxylate, by hepatic carboxylesterase (hCE). In patients infected with influenza, interleukin-6, a pro-inflammatory cytokine, is induced and this is known to reduce the expression of hCE in vitro.^34,35 Furthermore, inflammation and infection can alter the expression and activity of P-glycoprotein (P-gp) at the blood–brain barrier (BBB), down-regulating the activity of BBB P-gp after short-term exposure to inflammatory mediators.³⁶ It can be assumed that the activities of hCE and P-gp were reduced during the early phase or exacerbated state of influenza infection and that, subsequently, the concentration of unchanged OT may have increased in the plasma and brain.³⁷ As with the enhanced spontaneous behavioural activity following intraventricular administration of small doses of OT in mice,³⁸ the increased level of unchanged OT in human brains may be related to sudden-onset NPAEs occurring within 24 h of oseltamivir intake, which explains the high aOR of 1.91 observed in our 2 day time window analysis. In contrast, the mechanism of delayed onset-type NPAEs may be related to the inhibition of human endogenous neuraminidase activity by oseltamivir carboxylate,³⁹ although further studies are required.

Strengths and limitations

Our study has several strengths. First, use of a national sample cohort database can yield more representative results and overcome the limitations of previous studies that used only pharmacosurveillance reporting data or industry-funded clinical trials.^10,11,13 Furthermore, we effectively controlled for unmeasured confounders that were stable over time by applying a case-crossover design, which has been shown to minimize between-subject confounding and achieve optimal sample size.²⁰ We further adjusted for diagnosis of influenza and concomitant medication use potentially associated with the development of NPAEs.

Some potential limitations should be considered when interpreting our findings. First, as the measurements of NPAEs were based on claims data, there is a potential for inaccuracies in coding. However, we only included patients with the NPAEs recorded in the claims data, on the basis that the clinical condition of these patients may be more severe than that of those with no recorded NPAE, and there is no evidence that the distribution of oseltamivir exposure in the hazard and control periods would be different in these patients. While there are validity issues regarding diagnosis in all epidemiological studies using claims data, unlike the cohort design or case-control design, our case-crossover design does not result in bias caused by the unequal distribution of outcome misclassification. Therefore, we do not expect the main outcome to be incorrect. Second, exposure misclassification may have occurred between the hazard and control period. In practice, patients may take oseltamivir only when they have symptoms, although it is recommended that they complete the prescribed course of medication without discontinuing, even if symptoms begin to resolve during treatment. Therefore, exposure misclassification may have occurred due to the difference between the time frame of the prescribed regimen and the actual medication exposure. The risk may be overestimated if the exposure occurring in the control period was incorrectly measured as occurring in the hazard period and the risk may be underestimated in the reverse situation. Since exposure misclassification is more likely to occur with a shorter time window, we examined different time windows with a washout period to minimize the misclassification of exposure. Third, although the methodology and statistical analysis for this study removed the effect of time-invariant clinical characteristics, we could not identify residual confounding, due to the lack of information in the claims data, and there may have been differences in residual confounding between the hazard and control periods.

Conclusions

In summary, our study found an increased risk of NPAEs following oseltamivir use, irrespective of age, sex and presence or absence of medical conditions with a high risk for severe influenza complications. In the context of the increased use of oseltamivir globally, the benefits of oseltamivir should be carefully weighed against the potential risk of NPAEs.

Acknowledgements

This study used NHIS-NSC data (NHIS-20172-506).

Funding

This research was supported by Korea-Canada Cooperative Development Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (MSIT) (No. 2017K1A3A1A12073341).

Transparency declarations

None to declare.