Impact of long COVID phenotypes on quality of life following symptomatic omicron infection in Brazil: a machine learning analysis, 2025, Scolari et al

[Chandelier]

Impact of long COVID phenotypes on quality of life following symptomatic omicron infection in Brazil: a machine learning analysis

Spoiler: Authors

Scolari, Fernando Luis; Spinardi, Julia; Silva, Mariana Motta Dias da; Trott, Geraldine; Rodrigues, Cristina de Oliveira; Rover, Marciane Maria; Souza, Emanuel Maltempi de; Manfio, Josélia Larger; Camargo, Nathan Iori; Souza, Ana Paula de; Souza, Denise de; Carli, Raíne Fogliati De; Roldão, Emelyn de Souza; Mocellin, Duane; Miozzo, Aline Paula; Silva, Gabrielle Nunes da; Souza, Jennifer Menna Barreto de; Santos, Rosa da Rosa Minho dos; Itaqui, Carolina Rothmann; Rech, Gabriela Soares; Irineu, Vivian Menezes; Francisco, Saionara Cristina; Silvestre, Odilson; Neves, Precil Diego Miranda de Menezes; Tramujas, Lucas; Nobre, Vandack; Carvalho, Sidiclei Machado; Ferreira, Carlos Delmar do Amaral; Oliveira, Jaqueline Carvalho de; Royer, Carla Adriane; Luiz, Rafael Messias; Baura, Valter Antonio; Gradia, Daniela Fiori; Brandalize, Ana Paula Carneiro; Pereira, Hellen Abreu; Poitevin, Carolina Gracia; Robinson, Caroline Cabral; Barreto, Bruna Brandao; Schvartzman, Paulo R.; Marcolino, Milena; Antonio, Ana Carolina Peçanha; Polanczyk, Carisi Anne; Maccari, Juçara Gasparetto; Nasi, Luiz Antonio; Valluri, Srinivas Rao; Julião, Viviane Wal; d’Hellencourt, Florence Lefebvre; Kyaw, Moe H.; Castillo, Graciela Del Carmen Morales; Falavigna, Maicon; Rosa, Regis Goulart

Abstract
Background
This study aimed to identify phenotypes of long COVID symptoms in adults following Omicron infection and assess their association with health-related quality of life (HRQoL).

Methods
We analyzed three prospective observational studies in Brazil, enrolling adult patients who sought care for symptomatic Omicron infection between December 2021 and March 2023.
The infection was confirmed by either an antigen test or reverse transcriptase polymerase chain reaction.
Long COVID symptoms were assessed three months after enrollment through structured interviews.
Phenotypes of Long COVID-19 were identified using a machine learning-based clustering approach.
Exploratory analyses were conducted to examine predisposing factors and health-related quality of life utilities, measured by EQ-5D-3 L, associated with each phenotype.

Results
A total of 2,989 patients were analyzed (39% women, median age 41 years, and 96% had completed the primary series of COVID-19 vaccination).
Long COVID symptoms at three months were reported by 1,155 (38.6%) patients.
Three phenotypes were identified: cluster 1 (n = 459 [39.7%]), characterized by a median of three symptoms (IQR, 2–5) with memory loss (80.4%), concentration problems (38.3%) and fatigue (35.7%) being most common; cluster 2 (n = 549 [47.5%]), characterized by a median of two symptoms (IQR, 1–4) with fatigue (43.7%), other symptoms (42.3%), and cough (20.6%) being most common; and cluster 3 (n = 147, 12.7%), characterized by a higher number of symptoms (median, 8; IQR, 7–10), with fatigue (89.9%), memory loss (88.4%), and anxiety (64.6%) as the most common.
The mean EQ-5D-3 L utility at 3 months was 0.75 for cluster 1, 0.73 for cluster 2, and 0.59 for cluster 3 (p < 0.001).
After adjusted regression analysis, cluster 3 was independently associated with the lowest EQ-5D-3 L utilities (mean difference, -0.21; 95%CI, -0.24 to -0.18; p < 0.001).

Conclusions
Distinct phenotypic presentations of Long COVID following Omicron infection in Brazil were identified, with significant differences in quality of life.

Web | BMC Infectious Diseases
https://doi.org/10.1186/s12879-025-11956-6

 

[rvallee]

So, symptoms and their burden. As is always the case, but trying to over-complicate things in ways that add nothing. Those clusters don't mean a damn thing.

Not a single step beyond what was known on day 1: people experiencing symptoms that negatively impact their lives. We knew this. We still know this. Nothing done since then has clarified a single thing, in part because everyone is basically doing the same 3-4 studies with a local bent, despite the fact that it's entirely irrelevant.

 

[BrightCandle]

So, symptoms and their burden. As is always the case, but trying to over-complicate things in ways that add nothing. Those clusters don't mean a damn thing.

I recall RECOVER trying to do a similar thing and they largely ended up just shuffling the deck a little on the mild-very severe distinctions that the patients seemed to fall into. Number of symptoms or clustering around primary ones just ultimately ends up with lots of patients that match multiple clusters meaning they are useless as phenotypes. There is also a lot of AI clustering going on as well with complex dimensionality and that makes even less sense no one can work out how to use a measure that required 150 different random input blood parameters that don't seem to map usefully to the patients presenting symptoms.

 

[rvallee]

I recall RECOVER trying to do a similar thing and they largely ended up just shuffling the deck a little on the mild-very severe distinctions that the patients seemed to fall into. Number of symptoms or clustering around primary ones just ultimately ends up with lots of patients that match multiple clusters meaning they are useless as phenotypes. There is also a lot of AI clustering going on as well with complex dimensionality and that makes even less sense no one can work out how to use a measure that required 150 different random input blood parameters that don't seem to map usefully to the patients presenting symptoms.

Or the fact that symptoms fluctuate a lot over time. The only study I remember that did it correctly was the 2nd paper published by the Body Politic community, which tracked symptoms over time, showed how they can erratically come and go. It's been obvious for a long time now that this is not a useful thing to do.

I guess medicine is too used with symptoms being binary: there or not, and when they are there, they are permanent until treatment or death. They seem unable to unanchor from this. It sure doesn't help that symptoms fluctuations has been made a 'positive' feature of psychosomatic illness, and I really wonder how much good work is being avoided simply because some jerks sharpied "there be dragons" on a map and somehow people are following it like it's some kind of rule.