{"id":16147,"date":"2026-10-26T06:00:00","date_gmt":"2026-10-26T10:00:00","guid":{"rendered":"https:\/\/cov19longhaulfoundation.org\/?p=16147"},"modified":"2026-09-16T10:03:15","modified_gmt":"2026-09-16T14:03:15","slug":"long-covid-and-analysis-of-recovery-persistence-and-worsening-illnesses","status":"publish","type":"post","link":"https:\/\/cov19longhaulfoundation.org\/?p=16147","title":{"rendered":"Long COVID and Analysis of Recovery, Persistence and Worsening Illnesses"},"content":{"rendered":"\n<p class=\"has-small-font-size wp-block-paragraph\"><strong>John Murphy, CEO<\/strong>, <strong>The COVID-19 Long-haul Foundation<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Abstract<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Long COVID is increasingly recognized not as a uniform post-infectious syndrome but as a heterogeneous disease whose manifestations evolve over time. Yet the clinical and biological determinants of those trajectories remain incompletely characterized. The central question addressed in this analysis is: <strong>Can longitudinal clinical data distinguish patients who recover from those who remain persistently symptomatic, relapse intermittently, or develop worsening disease\u2014and can those trajectories eventually be predicted from information available early after infection?<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This article performs a secondary quantitative analysis of the published 2025 RECOVER-Adult trajectory study, which prospectively followed 3,659 participants from approximately three through 15 months after SARS-CoV-2 infection. Investigators used finite-mixture modeling of the Long COVID Research Index (LCRI) and identified eight longitudinal symptom profiles. We calculate the empirical proportions of each trajectory, approximate 95% confidence intervals, aggregate clinically meaningful trajectory categories, and examine the subgroup already meeting Long COVID criteria at three months.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The analysis finds that 638 of 3,659 participants\u2014<strong>17.4% (95% CI, 16.2\u201318.7%)<\/strong>\u2014belonged to the two groups characterized by persistent or intermittently high symptom burden. A further 526 participants\u2014<strong>14.4% (95% CI, 13.2\u201315.5%)<\/strong>\u2014belonged to the two worsening trajectories. Thus, nearly one-third of the cohort, <strong>31.8%<\/strong>, occupied trajectories characterized by either substantial persistent\/intermittent disease or worsening symptoms. Among the 377 participants meeting the Long COVID research threshold at three months, 175 (46.4%) followed a persistent-high trajectory, 132 (35.0%) an intermittently high trajectory, 66 (17.5%) an improving-moderate trajectory and only four (1.1%) an improving-low trajectory.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These results demonstrate that a single prevalence estimate substantially understates the importance of temporal disease behavior. The most consequential unresolved problem is therefore not merely determining how many individuals develop Long COVID, but determining whether early clinical, laboratory, physiological and electronic-health-record variables can predict an individual&#8217;s subsequent trajectory.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A definitive research program should integrate prospective patient-reported outcomes with electronic health records, claims data, laboratory testing, imaging, medication exposure, reinfection history, functional status and mortality over a minimum of five years.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Introduction<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The epidemiology of Long COVID has traditionally been framed around prevalence.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">How many people remain symptomatic after SARS-CoV-2 infection?<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That question was unavoidable during the emergence of the disease. But prevalence is a static measure imposed upon a dynamic illness.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A patient may improve steadily.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Another may remain chronically symptomatic.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A third may fluctuate between relative wellness and severe illness.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A fourth may initially appear well and subsequently deteriorate.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These are not interchangeable clinical phenomena.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The National Academies&#8217; working definition of Long COVID explicitly recognizes that the condition may be continuous, relapsing and remitting, or progressive. The natural history of the disease therefore requires longitudinal analysis rather than repeated cross-sectional prevalence surveys.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The 2025 RECOVER-Adult analysis represents a major advance because it followed 3,659 participants prospectively from infection through 15 months and used finite-mixture modeling to identify eight distinct symptom trajectories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The present article takes those published data one step further.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Rather than simply describing the eight profiles, we quantitatively examine their relative magnitude and ask what the observed distribution implies about the natural history of Long COVID.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Dataset<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The RECOVER-Adult trajectory analysis included <strong>3,659 participants<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Of these:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>3,280 were enrolled within 30 days of their first infection;<\/li>\n\n\n\n<li>379 were participants initially enrolled while uninfected who subsequently experienced their first infection;<\/li>\n\n\n\n<li>69% were female;<\/li>\n\n\n\n<li>99.6% were infected during the Omicron era;<\/li>\n\n\n\n<li>98% were not hospitalized during acute infection.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Participants completed symptom assessments at approximately 3, 6, 9, 12 and 15 months.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The investigators reported that 68% had data at all five visits, 18% had four visits, 9% had three visits and 4% had two visits. Missingness therefore exists and is an important consideration when interpreting the trajectory estimates.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The Long COVID Research Index (LCRI) was used as the longitudinal symptom measure, with an LCRI of <strong>11 or greater<\/strong> defining the study&#8217;s research threshold for Long COVID.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Quantitative Analysis of the Eight Trajectories<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The published cohort contained eight trajectory groups.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Table 1. Distribution of Long COVID trajectories<\/h5>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>Trajectory<\/th><th>n<\/th><th>Percentage of cohort<\/th><th>Approx. 95% CI<\/th><\/tr><tr><td>Persistent, high symptom burden<\/td><td>195<\/td><td><strong>5.33%<\/strong><\/td><td>4.60\u20136.06%<\/td><\/tr><tr><td>Intermittently high symptom burden<\/td><td>443<\/td><td><strong>12.11%<\/strong><\/td><td>11.05\u201313.16%<\/td><\/tr><tr><td>Improving, moderate symptom burden<\/td><td>379<\/td><td><strong>10.36%<\/strong><\/td><td>9.37\u201311.35%<\/td><\/tr><tr><td>Improving, low symptom burden<\/td><td>334<\/td><td><strong>9.13%<\/strong><\/td><td>8.19\u201310.06%<\/td><\/tr><tr><td>Worsening, moderate symptom burden<\/td><td>309<\/td><td><strong>8.44%<\/strong><\/td><td>7.54\u20139.35%<\/td><\/tr><tr><td>Delayed worsening<\/td><td>217<\/td><td><strong>5.93%<\/strong><\/td><td>5.17\u20136.70%<\/td><\/tr><tr><td>Consistent, low symptom burden<\/td><td>481<\/td><td><strong>13.15%<\/strong><\/td><td>12.05\u201314.24%<\/td><\/tr><tr><td>Consistent, minimal\/no symptom burden<\/td><td>1,301<\/td><td><strong>35.56%<\/strong><\/td><td>34.01\u201337.11%<\/td><\/tr><tr><td><strong>Total<\/strong><\/td><td><strong>3,659<\/strong><\/td><td><strong>100%<\/strong><\/td><td>\u2014<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Analysis:<\/strong> Percentages and confidence intervals are secondary calculations from the published trajectory counts. Confidence intervals use the conventional Wald approximation for a binomial proportion and are presented descriptively rather than as model-based uncertainty intervals.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The published trajectory counts sum exactly to the full analytic cohort:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>195 + 443 + 379 + 334 + 309 + 217 + 481 + 1,301 = 3,659.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That internal consistency provides a useful check on the quantitative reconstruction.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 1. Distribution of the Eight Trajectories<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>Percentage of RECOVER cohort\n\nMinimal\/no symptoms       \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 35.6%\nConsistent low            \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                       13.1%\nIntermittently high       \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                        12.1%\nImproving moderate        \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                          10.4%\nImproving low             \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                            9.1%\nWorsening moderate        \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                             8.4%\nDelayed worsening         \u2588\u2588\u2588\u2588\u2588\u2588                               5.9%\nPersistent high           \u2588\u2588\u2588\u2588\u2588                                5.3%<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">This distribution immediately reveals an important feature of Long COVID.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The cohort cannot reasonably be divided into two groups\u2014\u201cLong COVID\u201d and \u201cno Long COVID\u201d\u2014without losing substantial information.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">A More Informative Aggregation<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The eight categories can be grouped into clinically meaningful supercategories.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Group A: Persistent or intermittently high disease<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">195 + 443 = <strong>638 participants<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">638 \/ 3,659 = <strong>17.44%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Approximate 95% CI: <strong>16.21\u201318.67%<\/strong><\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Group B: Worsening disease<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">309 + 217 = <strong>526 participants<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">526 \/ 3,659 = <strong>14.38%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Approximate 95% CI: <strong>13.24\u201315.51%<\/strong><\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Group C: Improving trajectories<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">379 + 334 = <strong>713 participants<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">713 \/ 3,659 = <strong>19.49%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Approximate 95% CI: approximately <strong>18.20\u201320.78%<\/strong><\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Group D: Low or minimal symptom burden<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">481 + 1,301 = <strong>1,782 participants<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">1,782 \/ 3,659 = <strong>48.70%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Approximate 95% CI: approximately <strong>47.08\u201350.32%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These calculations produce a striking result.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 2. Four Clinically Meaningful Trajectory Categories<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>LOW \/ MINIMAL\n\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 48.7%\n\nIMPROVING\n\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                               19.5%\n\nPERSISTENT \/ INTERMITTENTLY HIGH\n\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                                 17.4%\n\nWORSENING\n\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                                    14.4%<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Almost half of participants followed consistently low or minimal-symptom trajectories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">But approximately <strong>51.3% did not<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The latter group includes patients with improving symptoms, persistent disease, intermittent disease and worsening disease.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This illustrates why a simple Long COVID prevalence estimate cannot adequately describe the longitudinal burden.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Most Important Quantitative Finding<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The most consequential calculation is the combination of the persistent\/intermittently high and worsening categories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">638 + 526 = <strong>1,164 participants<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">1,164 \/ 3,659 = <strong>31.81%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Thus, approximately <strong>one participant in three<\/strong> in this prospective cohort occupied a trajectory characterized either by substantial persistent\/intermittent symptom burden or by worsening symptoms during the observation period.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is not equivalent to saying that one-third of the population has Long COVID.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That would be an incorrect interpretation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Rather, it means that <strong>31.8% of this selected prospective cohort belonged to trajectory classes whose defining characteristics included persistent substantial symptoms, intermittent substantial symptoms, or worsening symptom burden.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That distinction is critical.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 3. The 31.8% High-Concern Trajectory Population<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>Entire RECOVER cohort: 3,659\n\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n\u2502                                                      \u2502\n\u2502                  OTHER TRAJECTORIES                  \u2502\n\u2502                         68.2%                        \u2502\n\u2502                                                      \u2502\n\u251c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n\u2502 Persistent\/intermittent      \u2502 Worsening             \u2502\n\u2502 high symptom burden          \u2502 trajectories          \u2502\n\u2502                              \u2502                       \u2502\n\u2502          17.4%               \u2502       14.4%           \u2502\n\u2502                              \u2502                       \u2502\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n\n                HIGH-CONCERN TRAJECTORIES\n                         31.8%<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">This secondary aggregation is particularly useful for future research because it provides a possible operational definition for a <strong>high-risk longitudinal phenotype<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It should not, however, be used clinically without prospective validation.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">What Happens to Patients Who Already Have Long COVID at Three Months?<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The most revealing analysis may be the subgroup analysis.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">At approximately three months, <strong>377 participants<\/strong> met the study&#8217;s LCRI threshold for Long COVID.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The subsequent trajectory was:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>Three-month Long COVID group<\/th><th>n<\/th><th>Percentage<\/th><\/tr><tr><td>Persistent, high<\/td><td>175<\/td><td><strong>46.4%<\/strong><\/td><\/tr><tr><td>Intermittently high<\/td><td>132<\/td><td><strong>35.0%<\/strong><\/td><\/tr><tr><td>Improving, moderate<\/td><td>66<\/td><td><strong>17.5%<\/strong><\/td><\/tr><tr><td>Improving, low<\/td><td>4<\/td><td><strong>1.1%<\/strong><\/td><\/tr><tr><td>Other trajectory categories<\/td><td>0<\/td><td><strong>0%<\/strong><\/td><\/tr><tr><td><strong>Total<\/strong><\/td><td><strong>377<\/strong><\/td><td><strong>100%<\/strong><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">These percentages are secondary calculations from the published counts. The underlying counts are reported by the RECOVER investigators.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 4. What Happened to the 377 Patients With Long COVID at Three Months?<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>Persistent high          \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 46.4%\n\nIntermittently high      \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588             35.0%\n\nImproving moderate       \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588                              17.5%\n\nImproving low            \u2588                                                   1.1%<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">This produces an important quantitative observation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Among people who had already crossed the Long COVID threshold at three months:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>81.4%<\/strong> subsequently belonged to either the persistent-high or intermittently-high groups.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Calculation:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">175 + 132 = 307<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">307 \/ 377 = <strong>81.43%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Approximate 95% CI: <strong>77.5\u201385.4%<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Conversely:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>18.6%<\/strong> belonged to improving trajectories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Thus, in this cohort, early established Long COVID was substantially more likely to persist or fluctuate than to rapidly resolve.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This does not prove that Long COVID is inevitably chronic.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It demonstrates that <strong>early established disease carries substantial probability of persistence<\/strong>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 5. Three-Month Prognostic Signal<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>Patients meeting Long COVID threshold at 3 months\n                         n = 377\n\n                    \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n                    \u2502               \u2502\n                    \u2502    81.4%      \u2502\n                    \u2502 Persistent or \u2502\n                    \u2502 intermittent  \u2502\n                    \u2502               \u2502\n                    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n                           \u2502\n                           \u2502\n                    \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n                    \u2502              \u2502\n                    \u25bc              \u25bc\n               46.4%            35.0%\n             Persistent       Intermittent\n\n\n                    18.6%\n                    \u2502\n          \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n          \u25bc                   \u25bc\n       17.5%                 1.1%\n      Improving             Improving\n      moderate                low<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">The three-month disease state therefore contains substantial prognostic information.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">But it is not sufficient to predict the entire future.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Three Months Versus Fifteen Months<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The investigators reported that 374 of 3,644 evaluable participants met the LCRI threshold at three months:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>10.26%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">At 15 months, 324 of 2,970 evaluable participants met the threshold:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>10.91%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Our calculations give approximate 95% confidence intervals of:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Three months: <strong>9.28\u201311.25%<\/strong><\/li>\n\n\n\n<li>Fifteen months: <strong>9.79\u201312.03%<\/strong><\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">At first glance, the prevalence appears remarkably similar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is an important result.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 6. Long COVID Research-Index Prevalence<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>LCRI \u2265 11\n\n3 months       10.26%   \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\n\n15 months      10.91%   \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">The difference is approximately:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>+0.65 percentage points<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">or approximately a <strong>6.3% relative increase<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">However, the important conclusion is not that prevalence increased.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The more important conclusion is that <strong>similar cross-sectional prevalence at two time points can conceal substantial movement of individuals between disease states.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A patient can recover while another develops new symptoms.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The aggregate prevalence may remain almost unchanged even though the underlying population is changing continuously.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is one of the central reasons longitudinal trajectory analysis is superior to prevalence alone.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Hidden Dynamics Behind Stable Prevalence<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Imagine 100 patients.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">If 10 have Long COVID at month three and 10 have Long COVID at month 15, one might conclude that nothing has changed.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That would be incorrect if:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>six of the original ten recovered;<\/li>\n\n\n\n<li>six previously unaffected patients developed Long COVID;<\/li>\n\n\n\n<li>four patients fluctuated;<\/li>\n\n\n\n<li>and four remained continuously symptomatic.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">The prevalence remains 10%.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The disease dynamics are completely different.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is a fundamental epidemiological principle:<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\"><strong>Stable prevalence does not imply stable disease.<\/strong><\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\">Long COVID therefore requires longitudinal transition analysis.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The RECOVER Data Demonstrate This Directly<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The RECOVER investigators reported that 36% of the full cohort experienced an on-study reinfection by 15 months, with reinfection proportions ranging approximately from 31% to 40% across the eight trajectory groups. The worsening trajectories had reinfection frequencies of approximately 39\u201340%.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is particularly interesting because the investigators found no obvious explanation whereby reinfection alone accounted for the increasing symptom burden in the worsening groups.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">They specifically noted that reinfection rates were comparable across trajectory profiles and suggested alternative possibilities including delayed pathophysiology or intercurrent illness.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That observation deserves considerably more investigation.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">A Potentially Important Statistical Signal<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The worsening groups contained:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>526 \/ 3,659 = 14.38%<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The two worsening groups were:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Worsening moderate: 8.44%<\/li>\n\n\n\n<li>Delayed worsening: 5.93%<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Their combined proportion therefore exceeds the persistent-high group by almost a factor of three.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">526 \/ 195 = <strong>2.70<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Thus, for every participant in the persistent-high group, approximately <strong>2.7 participants<\/strong> belonged to one of the two worsening groups.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This does not mean that worsening disease is biologically more important than persistent disease.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It does mean that <strong>worsening trajectories are too large to be dismissed as rare clinical anomalies<\/strong>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 7. Relative Size of Persistent and Worsening Phenotypes<\/h5>\n\n\n\n<pre class=\"wp-block-code\"><code>Persistent high          \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 195\n\nWorsening moderate       \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 309\n\nDelayed worsening        \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 217<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">The combined worsening population is therefore:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>526 versus 195<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">or approximately <strong>2.7 times larger<\/strong> than the persistent-high group.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That finding provides a strong rationale for dedicated investigation of progressive and delayed-onset Long COVID.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Unanswered Biological Question<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The statistical trajectories do not themselves establish mechanisms.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A worsening symptom trajectory could represent:<\/p>\n\n\n\n<ol start=\"1\" class=\"wp-block-list\">\n<li>delayed biological consequences of SARS-CoV-2 infection;<\/li>\n\n\n\n<li>autonomic dysfunction;<\/li>\n\n\n\n<li>immune dysregulation;<\/li>\n\n\n\n<li>persistent viral antigen or viral reservoirs;<\/li>\n\n\n\n<li>endothelial or vascular dysfunction;<\/li>\n\n\n\n<li>mitochondrial or metabolic abnormalities;<\/li>\n\n\n\n<li>neurological injury;<\/li>\n\n\n\n<li>an independent disease occurring after COVID-19;<\/li>\n\n\n\n<li>medication effects;<\/li>\n\n\n\n<li>ascertainment or reporting effects;<\/li>\n\n\n\n<li>some combination of these factors.<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">The RECOVER trajectory study did not have sufficient biological measurements to distinguish among these mechanisms. The investigators explicitly identified biological predictors and biomarkers of trajectory as an important direction for subsequent research.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This represents perhaps the most important gap in the current evidence.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The EHR Opportunity<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The United States now possesses an extraordinary longitudinal data resource.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">RECOVER reports access to electronic health records representing <strong>more than 60 million people<\/strong> through partnerships involving PCORnet and the National COVID Cohort Collaborative.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Separately, RECOVER\/N3C researchers have analyzed more than <strong>600 million condition diagnoses from approximately 14 million patients<\/strong>, using temporal topic modeling to identify clinical phenotypes associated with SARS-CoV-2 infection and Long COVID.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These datasets create the possibility of moving from symptom trajectories to <strong>multisystem disease trajectories<\/strong>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 8. The Proposed Integrated Dataset<\/h5>\n\n\n\n<pre class=\"wp-block-preformatted\"> <code>                        SARS-CoV-2 INFECTION\n                                  \u2502\n              \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n              \u25bc                   \u25bc                  \u25bc\n             EHR                CLAIMS             LABS\n              \u2502                   \u2502                  \u2502\n              \u251c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n              \u2502              \u2502                \u2502\n              \u25bc              \u25bc                \u25bc\n          Diagnoses       Utilization      Biomarkers\n              \u2502              \u2502                \u2502\n              \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n                             \u25bc\n                     PATIENT-REPORTED DATA\n                             \u2502\n                 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n                 \u25bc           \u25bc           \u25bc\n               Symptoms   Function     Quality of life\n                 \u2502           \u2502           \u2502\n                 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n                             \u25bc\n                        TRAJECTORY\n                             \u2502\n              \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n              \u25bc              \u25bc               \u25bc\n           Recovery      Persistence      Progression<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">The objective should be to link these data longitudinally at the individual level.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">A Five-Year Research Architecture<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The existing RECOVER study establishes the feasibility of intensive short-term trajectory measurement.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The next phase should extend follow-up substantially.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A proposed schedule would include:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><tbody><tr><th>Time<\/th><th>Principal measurements<\/th><\/tr><tr><td>Acute infection<\/td><td>severity, treatment, vaccination, laboratory values<\/td><\/tr><tr><td>30 days<\/td><td>symptoms, functional status, laboratory testing<\/td><\/tr><tr><td>3 months<\/td><td>Long COVID phenotype<\/td><\/tr><tr><td>6 months<\/td><td>symptoms + function + EHR<\/td><\/tr><tr><td>12 months<\/td><td>symptoms + function + organ-system evaluation<\/td><\/tr><tr><td>18 months<\/td><td>trajectory reassessment<\/td><\/tr><tr><td>24 months<\/td><td>comprehensive reassessment<\/td><\/tr><tr><td>36 months<\/td><td>organ-system outcomes<\/td><\/tr><tr><td>48 months<\/td><td>functional and biological status<\/td><\/tr><tr><td>60 months<\/td><td>durable recovery\/persistence\/progression<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">The most important addition would be repeated objective biological measurements.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">A Proposed Predictive Model<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The ultimate objective should be to determine whether early variables can predict trajectory.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">For example:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[<br>P(\\text{trajectory}_i | X_i)<br>]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">where:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>(X_i) represents the patient&#8217;s early clinical, biological and demographic features;<\/li>\n\n\n\n<li>and the trajectory could be recovery, persistence, intermittent disease, worsening or delayed worsening.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">A more complete model could be written:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[<br>P(T_i=t \\mid X_{i0},X_{i1},&#8230;,X_{ik})<br>]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">where (T_i) is the eventual trajectory and (X) represents observations collected over time.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The model should incorporate time-dependent variables rather than assuming that exposures remain constant.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Candidate Predictors<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The candidate predictor matrix should include:<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Demographic variables<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>age;<\/li>\n\n\n\n<li>sex;<\/li>\n\n\n\n<li>race and ethnicity;<\/li>\n\n\n\n<li>socioeconomic status;<\/li>\n\n\n\n<li>geographic location.<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">Pre-existing disease<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>cardiovascular disease;<\/li>\n\n\n\n<li>diabetes;<\/li>\n\n\n\n<li>kidney disease;<\/li>\n\n\n\n<li>pulmonary disease;<\/li>\n\n\n\n<li>autoimmune disease;<\/li>\n\n\n\n<li>neurological disease;<\/li>\n\n\n\n<li>obesity.<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">Acute COVID-19<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>hospitalization;<\/li>\n\n\n\n<li>oxygen requirement;<\/li>\n\n\n\n<li>ICU admission;<\/li>\n\n\n\n<li>antiviral treatment;<\/li>\n\n\n\n<li>corticosteroids;<\/li>\n\n\n\n<li>inflammatory markers.<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">Early Long COVID phenotype<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>fatigue;<\/li>\n\n\n\n<li>post-exertional malaise;<\/li>\n\n\n\n<li>cognitive dysfunction;<\/li>\n\n\n\n<li>sleep disturbance;<\/li>\n\n\n\n<li>dysautonomia;<\/li>\n\n\n\n<li>pain;<\/li>\n\n\n\n<li>sensory symptoms.<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">Objective measures<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>CBC;<\/li>\n\n\n\n<li>creatinine\/eGFR;<\/li>\n\n\n\n<li>glucose;<\/li>\n\n\n\n<li>HbA1c;<\/li>\n\n\n\n<li>inflammatory markers;<\/li>\n\n\n\n<li>cardiac biomarkers;<\/li>\n\n\n\n<li>pulmonary testing;<\/li>\n\n\n\n<li>imaging.<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">Subsequent exposures<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>reinfection;<\/li>\n\n\n\n<li>vaccination;<\/li>\n\n\n\n<li>medications;<\/li>\n\n\n\n<li>rehabilitation;<\/li>\n\n\n\n<li>major intercurrent disease.<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Critical Statistical Test<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">A clinically useful predictive model should be evaluated prospectively.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The following metrics should be reported:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Discrimination<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>AUROC;<\/li>\n\n\n\n<li>area under the precision-recall curve.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Calibration<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>calibration slope;<\/li>\n\n\n\n<li>calibration intercept;<\/li>\n\n\n\n<li>Brier score.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Clinical utility<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>decision-curve analysis;<\/li>\n\n\n\n<li>net benefit.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>External validity<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>independent geographic populations;<\/li>\n\n\n\n<li>different healthcare systems;<\/li>\n\n\n\n<li>different SARS-CoV-2 variant eras.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">A model that performs well internally but fails in another healthcare system should not be considered clinically validated.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Why Private Data Matter<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Publicly available studies are invaluable, but they cannot answer every question.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Private and restricted datasets can contain information that is rarely visible in conventional publications:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>detailed insurance claims;<\/li>\n\n\n\n<li>longitudinal pharmacy records;<\/li>\n\n\n\n<li>laboratory panels;<\/li>\n\n\n\n<li>specialist notes;<\/li>\n\n\n\n<li>disability claims;<\/li>\n\n\n\n<li>employment information;<\/li>\n\n\n\n<li>imaging;<\/li>\n\n\n\n<li>continuous monitoring;<\/li>\n\n\n\n<li>wearable-device data;<\/li>\n\n\n\n<li>rehabilitation records.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Appropriately de-identified and governed private datasets could therefore provide the missing link between symptoms and objective disease.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The crucial requirement is transparency.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Private data should be analyzed under protocols that specify:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>inclusion criteria;<\/li>\n\n\n\n<li>exclusion criteria;<\/li>\n\n\n\n<li>missing-data handling;<\/li>\n\n\n\n<li>prespecified outcomes;<\/li>\n\n\n\n<li>statistical methods;<\/li>\n\n\n\n<li>confounder adjustment;<\/li>\n\n\n\n<li>sensitivity analyses;<\/li>\n\n\n\n<li>validation procedures.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">The existence of proprietary data should never become a substitute for reproducibility.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Central Hypothesis<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The data support a testable hypothesis:<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\"><strong>Long COVID is not one longitudinal disease course but a family of partially overlapping trajectories whose future course can be predicted from combinations of early symptoms, host characteristics, acute disease severity, biological measurements and subsequent exposures.<\/strong><\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\">This hypothesis can now be tested.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The RECOVER trajectory data provide the phenotype.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The enormous EHR datasets provide the scale.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Biological cohorts provide mechanistic measurements.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Private claims and clinical datasets provide additional longitudinal resolution.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">What is missing is their systematic integration.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Figure 9. From Published Observation to Testable Prediction<\/h5>\n\n\n\n<pre class=\"wp-block-preformatted\"> <code>            RECOVER OBSERVATION\n                     \u2502\n                     \u25bc\n        Eight symptom trajectories\n                     \u2502\n                     \u25bc\n        \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n        \u2502      NEXT QUESTION      \u2502\n        \u2502                         \u2502\n        \u2502 What predicts trajectory\u2502\n        \u2502 before it develops?     \u2502\n        \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n                     \u2502\n        \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n        \u25bc            \u25bc            \u25bc\n       EHR         LABS       Patient data\n        \u2502            \u2502            \u2502\n        \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u253c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n                     \u25bc\n              Predictive model\n                     \u2502\n                     \u25bc\n             External validation\n                     \u2502\n                     \u25bc\n              Clinical trial\n                     \u2502\n                     \u25bc\n          Trajectory modification<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">The ultimate test is not prediction alone.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It is whether <strong>changing the predicted trajectory through treatment improves patient outcomes<\/strong>.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">What the Current Data Can and Cannot Tell Us<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The present evidence permits several conclusions.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">What can reasonably be concluded<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Long COVID has multiple longitudinal symptom trajectories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Persistent and intermittent high-burden trajectories constitute a substantial subgroup.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Worsening and delayed-worsening trajectories are also substantial.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Patients who meet the Long COVID threshold at three months are more likely to remain persistently or intermittently symptomatic than to rapidly improve.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Cross-sectional prevalence can conceal considerable movement between disease states.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Large EHR datasets make much larger longitudinal investigations feasible.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These conclusions are supported by the RECOVER data and related EHR research.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">What cannot yet be concluded<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The eight symptom trajectories are not proven biological subtypes.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Worsening symptoms do not automatically demonstrate progressive organ damage.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The RECOVER data do not establish that persistent virus causes any particular trajectory.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The data do not establish that reinfection causes worsening disease.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The trajectory model has not yet been demonstrated to predict five-year outcomes.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The relationship between symptom trajectory and objective organ dysfunction remains incompletely characterized.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These distinctions are essential to scientifically responsible interpretation.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Limitations of the Current Evidence<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The RECOVER trajectory cohort has several important limitations.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">First, 99.6% of participants were infected during the Omicron era.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Second, 98% did not require hospitalization during acute infection.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Third, follow-up in the published trajectory analysis extended only through 15 months.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Fourth, the primary trajectory measure was symptom-based rather than a comprehensive biological measure of organ-system disease.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Fifth, missing symptom data occurred at individual visits, with approximately 8\u201316% missingness across visits six through 15 months.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Sixth, the cohort was 69% female, which may influence generalizability.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Finally, finite-mixture models identify statistical patterns; they do not automatically establish biological mechanisms.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">A More Ambitious Definition of Recovery<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The data suggest that \u201crecovery\u201d should itself be reconsidered.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A patient should not necessarily be classified as recovered solely because a symptom score falls below a threshold.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Future studies should distinguish:<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Symptomatic recovery<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Symptoms substantially diminish.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Functional recovery<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Normal daily activities resume.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Biological recovery<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Objective abnormalities return toward baseline.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Durable recovery<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">Improvement persists over prolonged observation without substantial relapse.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Only the final state represents convincing evidence of durable resolution.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">The Five-Year Endpoint<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The most useful future endpoint would therefore not simply be:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Long COVID yes\/no.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Instead, investigators should classify participants at five years as:<\/p>\n\n\n\n<ol start=\"1\" class=\"wp-block-list\">\n<li><strong>Durable recovery<\/strong><\/li>\n\n\n\n<li><strong>Partial recovery<\/strong><\/li>\n\n\n\n<li><strong>Persistent stable disease<\/strong><\/li>\n\n\n\n<li><strong>Relapsing disease<\/strong><\/li>\n\n\n\n<li><strong>Progressive disease<\/strong><\/li>\n\n\n\n<li><strong>New late-onset disease<\/strong><\/li>\n\n\n\n<li><strong>Death or major disability<\/strong><\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">Such a classification would provide clinicians, patients and policymakers with information far more useful than a single prevalence estimate.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Conclusion<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">The published RECOVER data provide a powerful demonstration that Long COVID is temporally heterogeneous.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Among 3,659 prospectively followed participants, 195 (5.3%) followed a persistently high symptom trajectory, 443 (12.1%) an intermittently high trajectory, 309 (8.4%) a worsening trajectory and 217 (5.9%) a delayed-worsening trajectory. Our secondary aggregation of the published counts shows that <strong>17.4%<\/strong> occupied persistent\/intermittently high trajectories and <strong>14.4%<\/strong> occupied worsening trajectories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Among the 377 individuals meeting the Long COVID research threshold at three months, <strong>81.4% subsequently remained in persistent or intermittent high-burden trajectories<\/strong>, while 18.6% entered improving trajectories.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These numbers change the central research question.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The problem is no longer merely to establish that Long COVID persists.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The problem is to determine <strong>why individual patients follow different trajectories<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The United States now possesses the data infrastructure necessary to address this question. RECOVER has access to EHR information from more than 60 million people, while N3C\/RECOVER investigations have already demonstrated the ability to analyze hundreds of millions of longitudinal diagnoses across millions of patients.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The next step should be a coordinated five-year longitudinal analysis combining:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>symptoms + function + EHR + claims + laboratory data + imaging + treatment + reinfection + biological specimens.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The scientific objective should be prediction.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Can the future trajectory of a Long COVID patient be identified three months after infection?<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Can a patient at high risk for persistent or progressive disease be identified before disability becomes entrenched?<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Can biological markers distinguish patients destined to recover from those likely to remain ill?<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">And most importantly:<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\"><strong>Can an intervention change the trajectory once it has been predicted?<\/strong><\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\">If the answer to those questions becomes yes, Long COVID medicine will have moved beyond epidemiological description and into precision medicine.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The ultimate measure of success should not be how accurately researchers can count Long COVID.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It should be how accurately medicine can determine <strong>where the disease is going\u2014and whether that trajectory can be changed.<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">References<\/h5>\n\n\n\n<ol start=\"1\" class=\"wp-block-list\">\n<li>Thaweethai T, Donohue SE, Martin JN, et al. Long COVID trajectories in the prospectively followed RECOVER-Adult US cohort. <em>Nature Communications<\/em>. 2025;16:9557. doi:10.1038\/s41467-025-65239-4.<\/li>\n\n\n\n<li>National Academies of Sciences, Engineering, and Medicine. <em>A Long COVID Definition: A Chronic, Systemic Disease State with Profound Consequences<\/em>. Washington, DC: National Academies Press.<\/li>\n\n\n\n<li>RECOVER COVID Initiative. Electronic health records are an important tool to address Long COVID questions. October 1, 2025. RECOVER reports access to EHR data from more than 60 million people through PCORnet and N3C.<\/li>\n\n\n\n<li>O&#8217;Neil ST, Madlock-Brown C, Wilkins KJ, et al. Finding Long-COVID: Temporal topic modeling of electronic health records from the N3C and RECOVER programs. <em>npj Digital Medicine<\/em>. DOI:10.1038\/s41746-024-01286-3. The study analyzed more than 600 million condition diagnoses from approximately 14 million patients.<\/li>\n\n\n\n<li>Thaweethai T, et al. Supplementary statistical methods and trajectory modeling procedures. <em>Nature Communications<\/em>. 2025.<\/li>\n\n\n\n<li>Beale S, Yavlinsky A, Moncunill G, et al. Anti-nucleocapsid and anti-spike antibody trajectories in people with post-COVID condition versus acute-only infections: a nested longitudinal case-control study within the Virus Watch prospective cohort. <em>Nature Communications<\/em>. 2025.<\/li>\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h5 class=\"wp-block-heading\">Statistical Note<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\">All secondary percentages and confidence intervals presented in this article were calculated from the published RECOVER trajectory counts. They are not claimed to be reanalysis of the underlying individual-level RECOVER database. The confidence intervals are approximate binomial intervals calculated from published aggregate counts. The published investigators&#8217; finite-mixture model, multiple-imputation procedures and longitudinal modeling remain the authoritative statistical analysis of the original dataset.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">Conflict-of-Interest and Author Statement<\/h5>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>John Murphy is CEO of The COVID-19 Long-haul Foundation.<\/strong> The author advocates increased scientific investigation of Long COVID. The quantitative calculations in this article were performed from publicly reported aggregate data. No proprietary patient-level data were represented as independently analyzed. The article&#8217;s proposed predictive framework is a research hypothesis and should not be interpreted as a validated clinical prediction instrument.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>John Murphy, CEO, The COVID-19 Long-haul Foundation Abstract Long COVID is increasingly recognized not as a uniform post-infectious syndrome but as a heterogeneous disease whose manifestations evolve over time. Yet [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":16211,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1320,289,290,1527],"tags":[],"class_list":["post-16147","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-long-cov","category-long-haul-disease","category-long-term-effects","category-persistence"],"_links":{"self":[{"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/posts\/16147","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=16147"}],"version-history":[{"count":3,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/posts\/16147\/revisions"}],"predecessor-version":[{"id":16210,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/posts\/16147\/revisions\/16210"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=\/wp\/v2\/media\/16211"}],"wp:attachment":[{"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=16147"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=16147"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/cov19longhaulfoundation.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=16147"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}