In this study, we clustered Dutch patients suffering from PPCC into clinically distinct phenotypes. We discovered three phenotypes that show significant differences in terms of sex, age, symptom patterns, and impact on daily life. Fatigue was the most commonly reported symptom for the entire cohort, underlined by the average PROMIS Fatigue score being significantly higher than the average score for healthy Dutch children [22].

We found three distinct phenotypes of PPCC in our pediatric population, which is in line with a study conducted in adult patients with post-COVID-19 condition by Kenny et al.[17] This study reported one cluster with more cardiorespiratory symptoms, one cluster with more musculoskeletal pain, and one cluster with a significantly lower number of symptoms and burden of disease. This may suggest that disease presentation and underlying pathophysiology of post-COVID-19 condition could be similar in children and adults. However, other studies performing cluster analyses in adult populations with post-COVID-19 condition show contrasting results, identifying post-COVID-19 condition phenotypes with different characteristics [18, 29,30,31]. This might be due to differences in data collection methods (self-reported versus validated questionnaires) or variations in the duration since infection (long-term symptoms after > 4 weeks versus after > 12 months). However, it could also highlight the heterogeneity of the disease, emphasizing the necessity of independent validation of cluster analyses.

Two previously described risk factors for post-COVID-19 condition in adults and children are female sex [13, 18, 31] and older age [12]. In our cohort, we found similar characteristics to be potentially associated with a higher burden of disease, as evidenced by cluster 3 mainly consisting of younger boys who experienced the least symptoms and reported the lowest impact on daily life. Another previously described predictor for PPCC is severe acute COVID-19 with or without hospitalization during the acute phase [12, 32], but because our cohort only included children with mild acute clinical presentation of COVID-19, we could not investigate this risk factor in our study.

Post-COVID symptoms had a mild to severe self-reported impact on three domains of daily life (school, social interactions, and physical functioning) for almost all our participants. This is in line with findings in a cohort of Hungarian children with PPCC [33]. However, social restrictions and lockdowns during the pandemic can also affect mental and physical health [34] and must be taken into consideration when assessing the impact of PPCC symptoms on daily life.

PPCC is a heterogeneous disease, for which the identification of three distinct clinical phenotypes may advance our understanding of the underlying pathophysiological mechanisms of the disease. One hypothesis could be that autonomic dysfunction or viral reservoirs found in the brain might play a central role in cluster 1, characterized by symptoms such as dyspnea, exercise intolerance, and neurocognitive problems [15, 35, 36]. Another hypothesis for cardiorespiratory symptoms such as dyspnea and exercise intolerance could be persistent lung inflammation, as was recently described in an adult population with PCC [37]. Cluster 2 reported high numbers of gastrointestinal complaints such as abdominal pain, nausea, and loss of appetite, which could be a result of gut microbial dysbiosis, viral persistence, or altered neuro-immune interactions in the gut [15, 38]. In cluster 3, (recurrent) fever was a common symptom, which could be a result of autoimmunity, immune dysregulation due to chronic inflammation, or dysautonomia [15, 39]. Nevertheless, this cluster had a small sample size of only 16 patients, making it challenging to draw generalizable conclusions. All described theories need further in-depth investigations, preferably through (randomized) trials.

Our study has limitations. First, we only invited children who were referred to our tertiary care clinic to participate in our study. This means we only included patients with severe PPCC symptoms, which has created a selection bias, lowering the generalizability of our results. In addition, because this was a real-time study, there was no standard timeframe after infection that children were seen for the study visit, which explains the large variability in days since infection. Furthermore, because the POCOS study included children who were referred for standard care reasons, not all data was collected for every patient, which explains why there is missing data for some participants (e.g., the PROMIS Fatigue questionnaires). Second, the selection of the number of clusters was based on visual inspection of the clustering dendrogram instead of statistical methods. For the individual imputed datasets, the Dunn index [26] suggested a wide range for the number of clusters but selected three clusters most often. Finally, our cohort consisted of 111 patients, which is a relatively small sample size to perform cluster analyses on, hence why we were unable to perform validation analyses within this study cohort.

On the other hand, we were able to perform an unbiased hierarchical clustering analysis on a population with physician-diagnosed PPCC with mild acute SARS-CoV-2 infection, where alternative diagnoses were excluded. Another strength of our study lies in the scope of information we collected, providing a complete view of our participants, while biologic samples collected in the POCOS study allow for molecular characterization of the clusters in future analyses.

Classification of PPCC phenotypes can aid in comprehending its progression, identifying its causes, and ultimately developing management strategies tailored to specific phenotypes. In addition, the identification of phenotypes can help determine appropriate, personalized rehabilitation treatment strategies for children with PPCC.

Validation of these cluster analyses in a larger population is recommended to increase generalizability. Machine learning-based clustering has previously been applied to identify potential PPCC patients based on their clinical records [40], and has also been employed in the identification of PCC phenotypes in adults [30]. However, this method has not yet been performed for PPCC cluster analyses, making it a potentially promising tool. Further biomedical research, e.g., with a multi-omics approach, is needed to determine the possible underlying pathophysiology associated with these phenotypes.

In conclusion, PPCC is a heterogeneous and poorly characterized illness that can significantly affect the lives of children. This study found three distinct clinical phenotypes of PPCC that show differences in terms of gender, age, symptom patterns, and impact on daily life. These phenotypes may reflect different underlying pathophysiological mechanisms for post-COVID symptoms, which could help categorize patients for more successful monitoring and treatment strategies, as well as funnel future research into potential cluster targets.