The Association Between Suspected Long–COVID and Stimulant Prescribing in the United States

Abstract

Objectives:

This study sought to determine the association between suspected long–COVID and receipt of a stimulant prescription among persons diagnosed with COVID-19 and to describe clinical and demographic factors associated with receiving a stimulant prescription.

Methods:

US patients 18 and older who had a COVID–19 diagnosis or a positive COVID–19 PCR test from April 1^st, 2020 through December 21^st, 2022 recorded in a national electronic health record data set obtained from TriNetX were assessed. Comparison subjects were propensity score matched on baseline covariates to those with a symptom of or diagnosis of long–COVID. A Cox Proportional Hazards models was used to estimate the influence of long–COVID on stimulant prescription receipt.

Results:

Those with long–COVID (n=65,329) were twice as likely to be prescribed a stimulant as persons with only acute COVID–19 (n=189,438, HR=2.162; 1.929–2.423). Among persons with long–COVID, persons with new onset ADHD (HR=7.196; 5.749– 9.007), opioid-related disorders (HR=2.140; 1.264–3.621) and mood disorders (HR=1.649; 1.336–2.035) were more likely to be prescribed a stimulant.

Conclusion:

Further research describing the risks associated with increased stimulant use among persons with long–COVID is warranted.

1. Introduction

The COVID-19 pandemic has resulted in patients experiencing prolonged symptoms following acute COVID-19 infection, commonly referred to as long-COVID. Symptoms include, but are not limited to: respiratory failure, memory difficulties or “brain fog”, depression, stroke, dermatological changes, and other organ damage.^1,2 These symptoms can be persistent or transitory, and patients with severe primary COVID-19 infection, unvaccinated patients, older patients, or patients with multiple comorbidities are more likely to experience long-COVID symptoms.³ It is thought that longstanding symptoms following COVID-19 infection may be caused by a dysregulation of the immune system and may be similar in etiology to myalgic encephalomyelitis or chronic fatigue syndrome.⁴ Risk factors for the development of long-COVID include female gender, older age, severity of COVID-19 infection, high body mass index, and anxiety or depression diagnosis.^5,6 The lack of a homogenous definition of long-COVID definition has led to a wide range of long-COVID prevalence estimates. Rates of long-COVID have been estimated to be between 8% - 37% for patients with laboratory confirmed or suspected COVID-19 infection, but rates among hospitalized patients with COVID-19 are higher and may be as high as 68% based on patients in Wuhan, China.^7-9 The incidence of long-COVID is also influenced by COVID-19 variants; those infected with earlier COVID-19 variants have shown higher rates of long-COVID symptoms.¹⁰

Fatigue and memory problems are the most prevalent symptoms associated with long-COVID.^9,11-14 These symptoms have been found to impact patient quality of life, prompting the exploration of treatment strategies to alleviate these symptoms.¹⁵ The Centers for Disease Control (CDC) and the National Institute for Health and Care Excellence (NICE) guidelines for management of fatigue in long-COVID recommend non-pharmacologic therapies such as physical and occupational therapy and self-management of symptoms.^16,17 However, informal reports suggest some physicians are also prescribing stimulants, such as Adderall, for management of fatigue and brain fog associated with long-COVID.^18-20 Prescription stimulants are indicated for treatment of attention-deficit hyperactivity disorder (ADHD) and narcolepsy, but have been successfully used to treat depression-related fatigue.^21-24 The clinical utility of adding stimulants to manage long-COVID is currently unknown and must be weighed against recognized adverse effects, such as increased risk of heart arrhythmias, high blood pressure and the potential for substance abuse.²⁵ At the population level, the potential increase in stimulant prescribing for long-covid needs to be weighed against rise of overdose deaths attributed to stimulants that accompanied the COVID-19 pandemic.²⁶

This study sought to determine if and to what extent, persons with long-COVID or symptoms of long-COVID are more likely to be prescribed a stimulant than persons with prior COVID-19 infection without a long-COVID diagnosis or symptom using data from a nationally representative source of electronic health. Additionally, this study sought to determine the demographic and clinical characteristics, including the influence of individual long-COVID symptoms associated with being prescribed a stimulant.

2. Methods

2.1. Data Source

TriNetX, a globally operated, federated healthcare data repository of electronic health records that contains the clinical facts of over 250 million patients was used.²⁷ This dataset was derived from TriNetX’s Research network and contains patient demographic information, healthcare encounter information, diagnoses, procedures, medications, lab results, vital signs, cancer and cancer therapy information, and genomics results for all patients seen in contributing healthcare organizations.

2.2. Study Subjects

US residents aged 18 or older at the time of COVID-19 infection with a positive COVID-19 test (CPT-4 code 87635) or a COVID-19 diagnosis (ICD-10-CM U07.1) from April 1^st, 2020 through June 24^th, 2022 were included in the study. Patients with unknown patient geographic location, sex, or race were excluded. To ensure active engagement with the healthcare system during the study period, patients were required to have at least 2 clinical visits within the 6 months prior to their positive COVID-19 test. To detect incident stimulant use following COVID-19, patients were excluded if they had a prescription, administration, or hospital order for a stimulant in the 6 months prior to their positive COVID-19 test. Persons with a diagnosis for an absolute contraindication (cardiac arrhythmias, tachycardia, hypertensive urgency/emergency event, Tourette’s disorder, or bipolar type 1 disorder/manic depression) for stimulant use recorded in the 6 months prior to COVID were also excluded (supplementary table S1).²⁸ For the primary analytical cohort, patients were also excluded if they had one of the target symptoms or conditions associated with long-COVID (concentration and memory impairment, ADHD, weakness, malaise, or fatigue, depression) in the six month period prior to their COVID infection (supplementary table S2).

2.3. Study Design

Figure 1 displays the study design and time periods along with exclusion time periods. A 30 day waiting period starting with the date of the positive COVID-19 test, was required before a long-COVID diagnosis or symptom was considered. This was imposed to increase the confidence that the symptoms were associated with long-COVID and not a short term sequala of the acute COVID-19 infection or the recording of a comorbid condition. Starting with day 31 to day 180 after an acute COVID infection, patients were evaluated for the presence of either a ICD-10-CM diagnosis code for long-COVID (B94.8” prior to October 1^st, 2021 (a placeholder before an official long-COVID diagnosis code was released) and U09.9” from that date forward)²⁹ and ICD-10-CM diagnosis codes for symptoms and conditions associated with long-COVID for which a prescription stimulant could be reasonably prescribed (concentration and memory impairment, ADHD, weakness, malaise, or fatigue, depression; Supplementary Table S2) as determined by study personnel, from a more exhaustive list of long-COVID symptoms identified by Mizrahi, et. al.³⁰ The date of the earliest diagnosis for long-COVID or one of the target long-COVID symptoms was then set as the index date from which follow-up began. Those who did not have a long-COVID diagnosis code or target symptom or condition in day 31-180 made up the comparison cohort. Subjects in the comparison cohort had their index date set as the same day difference between date of COVID infection and the long-COVID diagnosis or associated symptom of their matched pair in the long-COVID cohort (described below). If a patient’s last record in the dataset precedes day 180 after the initial COVID diagnosis, that patient was excluded from the study to ensure patients were actively engaged with the health system throughout the entire long-COVID identification period.

Figure 1.

Study Diagram

2.4. Baseline Characteristics

Demographic characteristics collected include patient sex, race, age, and geographic region of the US (northeast, midwest, south, and west). Preexisting clinical characteristics were evaluated over the 180-day period before the index date. Additional clinical characteristics collected are described in the supplementary table (S3) and include potential contraindications to use (insomnia, anorexia, schizophrenia, schizotypical, or delusional disorders, mood disorders), adverse effects of prescription stimulants (substance use dependency) or are potential therapeutic uses (narcolepsy) and these were recorded as binary classification variables. The ICD-10-CM codes used to identify these baseline clinical characteristics are based upon the Agency for Healthcare Research Quality’s (AHRQ) Clinical Classification Software for ICD-10-CM.³¹

2.5. Matching

Long-COVID patients were matched at a 1:3 ratio based on baseline covariates using a greedy propensity score matching with a caliper set at 0.2 times the standard deviation of the logit of the propensity score with patients who had no evidence of long-COVID. Matching was evaluated using standardized differences calculated before and after the matching process. This was done to create comparable periods between an acute COVID-19 infection and the beginning of the evaluation periods.

2.6. Outcome Definition and Censoring

Patients were followed from their index date until they had a stimulant prescribed, defined as a prescription, administration, or hospital order for a stimulant, or were censored if they had a second laboratory confirmed COVID-19 infection, the end of the study period occurred, or until their last record in the TriNetX database. Stimulants included one or more RxNorm codes for the following stimulants: methylphenidate, dexmethylphenidate, amphetamine/dextroamphetamine mixture, dextroamphetamine, amphetamine, lisdexamfetamine, or serdexmethylphenidate/dexmethylphenidate were included, along with wakefulness promoting agents modafinil and armodafinil (supplement S4).

2.7. Statistical Analysis

The primary statistical approach was a time-to-event analysis using a Cox proportional hazards model to estimate a hazard of prescription stimulant use between the long-COVID cohort and comparison cohort. Since all potentially confounding covariates available were included in the propensity score model, no additional covariates were included in this model. To determine the factors associated with receiving a stimulant among long-COVID patients, a second Cox proportional hazards model was estimated using only the long-COVID cohort. This model included all the covariates previously described (all baseline variables, table S3) and additionally, included the indexing event (long-COVID diagnosis or associated long-COVID symptoms, in non-exclusive binary categories, table S2) as covariates in the model. All statistical analyses and data manipulation were done using SAS Enterprise Guide 8.3.

2.8. Sensitivity Analyses

A range of sensitivity analyses were performed to assess the influence of key study decisions on the findings. First, a sensitivity analysis permitted persons to have a long-COVID diagnosis, not including target symptoms, during the 30 day waiting period after an initial COVID-19 diagnosis or test result. These patients were excluded in the primary analysis as long-COVID diagnosis within 30 days of COVID-19 diagnosis raised concerns that the COVID-19 test would have been taken outside of the health system (I.e. at home), thus giving us a COVID-19 diagnosis date that may have occurred weeks after testing positive as this time was when it was diagnosed by a provider and present in our dataset. Because a diagnosis of ADHD is an indication for receiving a stimulant and was found to be a significant driver stimulant receipt, several additional analyses were conducted. Three models were estimated; one excluding anyone with an ADHD diagnosis from the start of the baseline period to the end of the exposure period, one excluding persons less than 25 year of age, and a third excluding persons under 35. The posteriori assumption leading the exclusion of younger age groups was that new onset ADHD diagnosis in older age groups is much less likely, thus lessening the risk of true, new onset ADHD diagnoses impacting our findings. Additionally, a model was estimated using a cohort of long-COVID patients that were identified using only a long-COVID diagnosis code and did not consider an associated symptom to define long-COVID.

3. Results

3.1. Patient Characteristics and Matching

Prior to propensity score matching 65,329 long-COVID subjects and 588,759 comparison subjects were retained after applying the study inclusions and exclusion criteria (Figure 2). After propensity score matching , 56,273 long-COVID patients were matched to 189,438 patients in the comparison cohort for the primary analysis. The demographic and clinical characteristics of these patients before and after propensity score matching can be found in Table 1. Prior to propensity score matching Long-COVID patients were older (long-COVID age 66-89, 23.64%; Comparison age 66-89, 18.36%), proportionally more female (long-COVID 65.88%; Comparison 60.77%) and white (long-COVID 78.94%; Comparison 73.22%), and were more likely to have baseline cancer (long-COVID 16.89%; Comparison 9.28%), mood disorders (long-COVID 9.58%; Comparison 2.87%), insomnia (long-COVID 5.10%; Comparison 2.45%), and nicotine related disorders (long-COVID 8.18%; Comparison 5.63%). The covariate distribution between long-COVID and comparison subjects were well balanced after propensity score matching and applying the final exclusion criteria as evidenced by all standardized differences of less than 0.10. For the secondary analysis to explore factors associated with receiving a stimulant prescription, 65,329 long-COVID patients were retained from the primary cohort prior to matching and second round exclusions.

Figure 2.

Patient Sample Derivation from Inclusion/Exclusion

* Counts are in non-exclusive categories.

** The second exclusion round was used to exclude comorbidities and stimulant use in the time period between COVID-19 diagnosis and long-COVID diagnosis.

⁺ ADHD: Attention-Deficit Hyperactivity Disorder

Table 1.

Baseline Characteristics Before and After Propensity Score Matching

	Unadjusted					After Propensity Score Matching*
	Long COVID Cohort		Comparison Cohort		Standardized Difference+	Long COVID Cohort		Comparison Cohort		Standardized Difference+
	N 65,329	%	N 588,759	%		N 56,273	%	N 189,438	%
Patient Age
18-25	5,705	8.73	62,520	10.62	−0.06	4,822	8.57	16,396	8.66	0.00
26-35	9,091	13.92	98,490	16.73	−0.08	7,774	13.81	26,059	13.76	0.00
36-45	9,696	14.84	95,016	16.14	−0.04	8,408	14.94	28,200	14.89	0.00
46-55	11,364	17.40	103,765	17.62	−0.01	10,009	17.79	33,161	17.50	0.01
56-65	13,131	20.10	116,511	19.79	0.01	11,482	20.40	38,231	20.18	0.01
66-89	15,445	23.64	108,117	18.36	0.13	13,057	23.20	44,706	23.60	−0.01
90 and over	897	1.37	4,340	0.74	0.06	721	1.28	2,685	1.42	−0.01
Sex
Female	43,041	65.88	357,761	60.77	0.11	37,390	66.44	125,620	66.31	0.00
Race
American Indian or Alaska Native	324	0.50	2,544	0.43	0.01	285	0.51	884	0.47	0.01
Asian	1,349	2.06	16,143	2.74	−0.04	1,184	2.10	3,888	2.05	0.00
Black or African American	12,009	18.38	138,040	23.45	−0.12	10,504	18.67	34,347	18.13	0.01
Native Hawaiian or Other Pacific Islander	76	0.12	972	0.17	−0.01	66	0.12	182	0.10	0.01
White	51,571	78.94	431,060	73.22	0.13	44,234	78.61	150,137	79.25	−0.02
Patient Location (Census Region)
East North Central	14,296	21.88	125,640	21.34	0.01	12,659	22.50	41,298	21.80	0.02
Middle Atlantic	16,331	25.00	165,231	28.06	−0.07	14,275	25.37	47,808	25.24	0.00
Mountain	2,645	4.05	23,235	3.95	0.01	2,281	4.05	7,773	4.10	0.00
New England	1,870	2.86	23,636	4.01	−0.06	1,742	3.10	5,574	2.94	0.01
Pacific	651	1.00	3,050	0.52	0.06	586	1.04	1,906	1.01	0.00
South Atlantic	14,167	21.69	107,505	18.26	0.09	11,849	21.06	40,992	21.64	−0.01
West North Central	8,358	12.79	35,048	5.95	0.24	6,710	11.92	24,409	12.88	−0.03
West South Central	7,011	10.73	105,414	17.90	−0.21	6,171	10.97	19,678	10.39	0.02
Match Date (Quarter)
2020Q2	7,709	11.80	81,425	13.83	−0.06	6,707	11.92	22,648	11.96	0.00
2020Q3	8,441	12.92	74,950	12.73	0.01	7,152	12.71	24,178	12.76	0.00
2020Q4	13,002	19.90	115,839	19.68	0.01	11,077	19.68	37,020	19.54	0.00
2021Q1	9,641	14.76	84,870	14.42	0.01	8,277	14.71	27,263	14.39	0.01
2021Q2	4,715	7.22	42,300	7.18	0.00	4,092	7.27	13,296	7.02	0.01
2021Q3	6,004	9.19	56,358	9.57	−0.01	5,248	9.33	17,967	9.48	−0.01
2021Q4	7,617	11.66	67,982	11.55	0.00	6,603	11.73	22,704	11.98	−0.01
2022Q1	6,563	10.05	51,821	8.80	0.04	5,668	10.07	19,411	10.25	−0.01
2022Q2	1,637	2.51	13,214	2.24	0.02	1,449	2.57	4,951	2.61	0.00
Anorexia Nervosa
	29	0.04	93	0.02	0.01	23	0.04	56	0.03	0.01
Insomnia
	3,331	5.10	14,401	2.45	0.14	2,860	5.08	8,937	4.72	0.02
Narcolepsy
	392	0.60	1,793	0.30	0.05	317	0.56	927	0.49	0.01
HIV **
	904	1.38	5,814	0.99	0.04	762	1.35	2,403	1.27	0.01
Cancer (Includes in situ carcinoma)
	11,034	16.89	54,665	9.28	0.23	9,064	16.11	31,253	16.50	−0.01
Delusional Disorders
	808	1.24	4,025	0.68	0.06	650	1.16	1,762	0.93	0.02
Mood Disorders
	6,260	9.58	16,901	2.87	0.29	5,497	9.77	14,937	7.88	0.07
Nicotine Related Disorders
	5,345	8.18	33,169	5.63	0.10	4,467	7.94	14,897	7.86	0.00
Alcohol Related Disorders
	1,374	2.10	7,287	1.24	0.07	1,118	1.99	3,507	1.85	0.01
Opioid Related Disorders
	834	1.28	3,902	0.66	0.06	685	1.22	2,072	1.09	0.01
Sedative Related Disorders
	103	0.16	375	0.06	0.03	92	0.16	231	0.12	0.01
Stimulant Related Disorders
	415	0.64	2,349	0.40	0.03	316	0.56	953	0.50	0.01
Cannabis Related Disorders
	576	0.88	3,790	0.64	0.03	484	0.86	1,283	0.68	0.02
Inhalant Related Disorders
	3	0.00	15	0.00	0.00	2	0.00	9	0.00	0.00
Hallucinogen Related Disorders
	42	0.06	183	0.03	0.01	32	0.06	89	0.05	0.00
Other Specified Substance Use Disorders
	1,139	1.74	5,535	0.94	0.07	909	1.62	2,697	1.42	0.02

^*Patients matched on all covariates listed in Table 1.

⁺Calculated using Cohen’s h formula.

^**HIV: Human Immundodeficiency Virus

3.2. Primary Analysis

790 patients (1.40%) in the long-COVID cohort received one or more prescription orders for a stimulant compared to 1,266 (0.67%) in the comparison cohort (Table 2). Patients in the comparison cohort had a longer average follow-up time compared to the long-COVID cohort (337 days vs 276 days). The most commonly used prescription stimulants were amphetamines (200, 0.36%, including amphetamine/dextroamphetamine salts), methylphenidate (193, 0.34%), dextroamphetamine (142, 0.25%), and lisdexamphetamine (107, 0.19%). Only 4.73% of the long-COVID cohort had one of the two diagnoses codes specific for long-COVID recorded. The most common long-COVID symptom defining the index diagnosis was malaise/fatigue (52.88%), followed by depression (34.60%), and with concentration/memory impairment (17.83%) while only 2.11% of the cohort had an ADHD diagnosis (data not shown).

Table 2.

Stimulant Prescribing in Long-COVID and Comparison Cohorts

	Long-COVID Cohort		Comparison Cohort
	N 56,273	%	N 189,438	%
Total (Any Stimulant)	790	1.40	1,266	0.67
Amphetamine*	200	0.36	352	0.19
Armodafinil	13	0.02	21	0.01
Dexmethylphenidate	13	0.02	20	0.01
Dextroamphetamine	142	0.25	221	0.12
Lisdexamphetamine	107	0.19	196	0.10
Methamphetamine	31	0.06	50	0.03
Methylphenidate	193	0.34	291	0.15
Modafinil	91	0.16	115	0.06

^*Includes Dextroamphetamine/Amphetamine salts

Persons with long-COVID were approximately twice as likely to be prescribed a stimulant than persons with only persons with acute COVID-19 (Table 3: HR=2.162; 95%CI: 1.929 – 2.423). A visual examination of the survival function and log(−log(survival)) vs log(time) graph did not suggest a violation of the proportional hazards assumption (supplement S5).

Table 3.

Hazard Ratios for Receipt of a Stimulant Prescription of Primary Analysis and Sensitivity Analyses

	N (LC, controls)		Hazard Ratio	95% CI
Primary Analysis	56,273	189,438	2.162	1.929	2.423
Age over 25	51,451	172,869	2.149	1.895	2.438
Age over 35	43,676	146,344	1.943	1.661	2.274
LC Code only	2,881	9,363	1.209	0.553	2.645
No exclusion for long-COVID diagnosis during waiting period	57,708	193,851	2.250	2.007	2.522
ADHD Excluded Day - 180 -180	55,514	185,896	1.824	1.616	2.059

Abbreviations: LC: Long-COVID, 95%CI: 95% Confidence Interval, ADHD: Attention Deficit Hyperactivity Disorder

3.3. Primary Analysis – Sensitivity Analyses

Sensitivity analyses exploring the veracity of the findings with different study design approaches and inclusion criteria showed similar results as the primary analysis (Table 3). Excluding those with new onset ADHD diagnosis, and excluding ADHD as a long-COVID symptom, resulted in a slightly lower risk of stimulant prescribing associated with long-COVID compared to the primary analysis (HR=1.82, 95%CI: 1.616 – 2.059). Excluding those under the age of 25 (HR=2.149, 95%CI: 1.895 – 2.438 ) and excluding those under the age of 35 (HR=1.943, 95%CI: 1.661 – 2.274) resulted in a slightly lower risk of stimulant prescribing compared to the primary analysis. Removing the exclusion for having a long-COVID diagnosis code during the 30 day waiting period resulted in largely unchanged results (HR=2.250, 95%CI: 2.007 – 2.522). When the analysis was restricted to include only those with a long-COVID diagnosis code in the long-COVID cohort, long-COVID subjects no longer showed a significant increase in stimulant receipt (HR=1.209, 95%CI: 0.553 – 2.645).

3.4. Secondary Analysis

In the secondary analysis we explored the association between patient demographics, baseline diagnoses, and long-COVID diagnoses and stimulant use among persons with long-COVID. An ADHD diagnosis as the index diagnosis had the strongest association with a prescription for a stimulant (Table 4: HR=7.196, 95%CI: 5.749 – 9.007). As age increased, the likelihood that a stimulant would be prescribed decreased. Asians (HR=0.471, 95%CI: 0.254 – 0.888) and Blacks (HR=0.417, 95%CI: 0.321 – 0.541) were less likely to be prescribed a stimulant compared to whites. Subjects with a baseline diagnosis for narcolepsy (HR=3.337, 95%CI: 1.915 – 5.815), mood disorders (HR=1.649, 95%CI: 1.336 – 2.035), and opioid related disorders (HR=2.140, 95%CI: 1.264 – 3.621) were more likely to be prescribed a stimulant.

Table 4.

Maximum Likelihood Estimates of Covariate Effects on Stimulant Receipt Among Long-COVID Cohort

Parameter		Hazard Ratio	95% Confidence Limits
Long-COVID Code		0.679	0.392	1.177
Depression Code		0.804	0.657	0.985
ADHD* Code		7.196	5.749	9.007
Malaise and Fatigue Code		0.711	0.581	0.871
Concentration and Memory Impairment Code		1.126	0.913	1.390
Patient Age (Ref: 90 +)
	18-25	7.023	1.727	28.548
	26-35	8.253	2.041	33.376
	36-45	5.679	1.400	23.025
	46-55	4.064	1.001	16.492
	56-65	2.120	0.518	8.668
	66-89	1.613	0.393	6.613
Patient Regional Location (Ref: West South Central)
	East North Central	0.708	0.520	0.965
	Middle Atlantic	1.178	0.880	1.576
	Mountain	0.790	0.506	1.235
	New England	1.777	1.162	2.719
	Pacific	1.890	0.898	3.975
	South Atlantic	1.096	0.803	1.497
	West North Central	1.630	1.204	2.207
Sex (Ref: Male)
	Female	1.104	0.939	1.297
	Asian	0.474	0.254	0.888
	Black or African American	0.417	0.321	0.541
	Pacific Islander/Native American	0.613	0.196	1.917
Anorexia		1.487	0.363	6.095
Insomnia		1.021	0.727	1.433
Narcolepsy		3.337	1.915	5.815
HIV*		1.012	0.554	1.849
Cancer (including in situ carcinoma)		1.123	0.896	1.408
Delusional Disorders		0.741	0.380	1.447
Mood Disorders		1.649	1.336	2.035
Alcohol Related Disorders		1.045	0.603	1.811
Opioid Related Disorders		2.140	1.264	3.621
Sedative Related Disorders		2.205	0.748	6.504
Stimulant Related Disorders		1.075	0.402	2.871
Nicotine Related Disorders		0.825	0.615	1.106
Cannabis Related Disorders		0.404	0.145	1.126
Hallucinogen and Other Related Disorders		0.663	0.358	1.226

^**Abbreviations. ADHD: Attention-Deficit Hyperactivity Disorder HIV: Human Immundodeficiency Virus

4. Discussion

Among a representative sample of persons seeking care in the U.S., persons with a long-COVID diagnosis or a long-COVID symptom for which a stimulant might be considered, about 1.5% of patients were prescribed a stimulant following their long-COVID diagnosis or symptom. This stimulant prescribing rate is more than double the rate observed among COVID-19 patients without long-COVID after balancing baseline characteristics between long-COVID and non long-COVID patients through propensity score matching. Applying this increased stimulant use rate to the estimated 7.5% of adults reporting long-COVID symptoms, as many as 274,000 American adults may have been prescribed a stimulant after their long-COVID symptom and half of this estimate would be attributed to the excess stimulant prescribing rate associated with long-COVID.³² Stimulants were more likely to be prescribed in younger adults, whites, those with ADHD as their long-COVID condition and those with preexisting mood disorders and opioid related disorders. These data suggest that clinicians may be experimenting with stimulants to help manage the symptoms of long-COVID particularly for younger white patients and those that may be more vulnerable to misuse or abuse.

Not surprisingly, persons with new onset ADHD as their long-COVID symptom were the ones most likely to be prescribed a stimulant. Analysis of prescribing trends in Costa Rica and Italy showed that the COVID-19 pandemic led to sharp increases in methylphenidate and atomoxetine, 2 commonly used drugs for ADHD, use from 2019 – 2022, suggesting increased ADHD treatment over this time period.³³ We have not found a study directly linking long-COIVD and new onset ADHD; however, ADHD diagnosis code F90.0 has been used to identify concentration/memory impairment in post-COVID patients.³⁰ Stimulants are the preferred pharmacotherapy for adult ADHD and risks of addiction are lower in ADHD patients treated with stimulants than ADHD patients without treatment.^34,35 Given the well-established efficacy of stimulants for ADHD and the unclear relationship between long-COVID and new onset ADHD, we excluded ADHD as one of the long-COVID target symptoms and still found that persons with long-COVID were nearly twice as likely to be prescribed a stimulant, suggesting that stimulants are being prescribed for other symptoms associated with long-COVID.

Long-COVID is estimated to affect 7% – 41% of non-hospitalized adults with an acute COVID-19 infection.³⁶ Fatigue and cognitive impairment are the most commonly reported symptoms of long-COVID.^9,11-14 The high prevalence of these long-COVID symptoms combined with our findings highlight the need for further research on the safety and efficacy of prescription stimulants when used for treatment of long-COVID symptoms. To our knowledge, there is no published evidence evaluating stimulants for symptoms associated with long-COVID. Our data suggest that persons with mood disorders, including depression were more likely to be prescribed a stimulant. A systemic review of clinical trials on the use of stimulants for depression found modafinil, armodafinil, dextroamphetamine, and methylphenidate to have adequate evidence to support improvement of depression related fatigue and cognitive impairment.²¹ Long-COVID-related fatigue, however, has been associated with postural orthostatic tachycardia syndrome (POTS), which can cause both rapid heartrate and severe fatigue.³⁷ Stimulant use for long-COVID related fatigue may have less efficacy than when used for depression related fatigue due to the different pathophysiological causes for the fatigue, and may carry greater cardiovascular risks since both POTS and stimulants can cause elevated heart rate. Stimulant trials for POTS treatment outside of long-COVID have shown some positive impact but are underpowered to determine if benefits of use exceed risks.³⁸ Cognitive impairment present in long-COVID does not yet have a well-established cause but has been hypothesized to occur due to brain damage either directly from the COVID-19 virus itself or through hypoxic changes resulting from respiratory distress in COVID-19 patients.³⁹ This pathology may be similar to traumatic brain injury, for which modafinil has been trialed but did not show a difference vs placebo, or to age-related dementias, for which methylphenidate and amphetamine have shown limited efficacy.^40,41 Ultimately, more research is going to be needed to justify use of stimulants for long-COVID and ensure we are giving more benefit than harm.

It is concerning that persons with prior opioid-related disorders were approximately twice as likely to receive a stimulant than persons without prior-opioid abuse, as concurrent use of opioids and stimulants has been estimated to double the risk of overdose.⁴² In 2021, 65.7% of fatal stimulant (excluding cocaine) overdoses also involved an opioid, although this category is dominated by methamphetamine overdoses and not prescribed stimulants.⁴³ Stimulant treatment reduces risks of substance use diagnosis among ADHD patients.³⁵ However, long-COVID patients may be at greater risk of stimulant misuse due to the mental health effects that accompanying the disease, which include depression, mood swings, anxiety, and psychotic disorders.⁴⁴ The Substance Abuse and Mental Health Services Administration included the need for identification and treatment of substance abuse disorders in long-COVID patients in recent guidelines.⁴⁵ With approximately 0.5% of the US population having a prescription stimulant use disorder in 2021, further research is warranted to assess the impact of this increased stimulant use in long-COVID patients on the health of US residents.⁴⁶

4.1. Limitations

Several limitations exist for our study. First, despite using propensity score matching on a range of potential confounding variables, unmeasured confounding could distort the strength and potentially the direction of the reported associations. Second, electronic health records often do not paint a complete picture of a patient’s medical history due to uncertainty in completeness of records as patients could seek care outside of the contributing healthcare organization, which can result in misclassification bias as important patient variables are not present in the dataset. We have attempted to minimize the impact of the limitation excluding persons without 2 clinical visits to the health system in the preceding 6 months, however, this potentially limits generalizability of these findings to patients that were actively engaged with their health system prior to being diagnosed with COVID-19. Third, TriNetX reports limited medication information, giving only an RXCUI for identifying the medication and not consistently specifying details of the medication use such as dose, strength, specific product used, and whether it is a prescription vs given in an inpatient setting. It is likely the majority of stimulants observed in our study are prescriptions, not inpatient administrations, since FDA approved indications for stimulants include ADHD, binge-eating disorder, and narcolepsy are primarily treated in the outpatient setting.⁴⁷ Fourth, we defined long-COVID using ICD-10-CM diagnosis codes of previously identified symptoms of long-COVID in addition to long-COVID diagnoses, however, the novelty of long-COVID means that currently there is not a well-defined and validated outcome definition for identifying these patients using ICD-10 diagnosis codes alone. Further, only 4.73% of patients identified as long-COVID patients in this study received one of the two diagnoses specific to long-COVID and this relatively low rate of specific long-COVID diagnoses is consistent with other studies using similar data.⁴⁸ Related, the symptoms used to identify long-COVID in this study were limited to CNS symptoms and not all long-COVID symptoms, such as respiratory symptoms, and therefore the rates of stimulant prescribing only apply to those with a long-COVID diagnosis and those with a CNS symptom for which a stimulant is more likely to be prescribed. Given the lack of a consensus definition for long-COVID symptoms combined with the fact that an overwhelming majority of long-COVID patients identified in our study were identified by symptoms and not an actual long-COVID diagnosis diminishes the confidence in defining long-COVID in this study. Fortunately, when we varied our long-COVID definition to probe the impact of alternative definitions of long-COVID on our study findings, we get similar results.

4.2. Conclusions

Stimulant prescribing among a cohort of long-COVID patients, identified through long-COVID diagnosis or diagnosis of long-COVID symptoms at least 30 days after acute infection, was more than double the rate when compared to COVID-19 patients without long-COVID symptoms. The efficacy and safety of stimulant treatment for long-COVID symptoms has not been established despite accompanying cardiorespiratory symptoms of long-COVID that could interact with stimulant use. Additionally, long-COVID patients with prior opioid-related disorders were over twice as likely to be prescribed a stimulant, which is concerning given the sharp rise in stimulant and opioid polysubstance abuse after the start of the pandemic. Further research into prescribing characteristics, safety and efficacy, and long-term outcomes of stimulant use for long-COVID symptoms is warranted.

Highlights.

• We found increased stimulant prescribing among long-COVID patients with fatigue and concentration issues.

• Stimulant use for this population has been untested and it is unknown if benefits outweigh risks.

• More research is needed to determine if this increased prescribing is increasing risks of adverse events related to stimulant abuse.