Background The ability to accurately predict early progression of dengue to severe disease is crucial for patient triage and clinical management. Previous systematic reviews and meta-analyses have found significant heterogeneity in predictors of severe disease due to large variation in these factors during the time course of the illness. We aimed to identify factors associated with progression to severe dengue disease that are detectable specifically in the febrile phase.Methods We did a systematic review and meta-analysis to identify predictors identifiable during the febrile phase associated with progression to severe disease defined according to WHO criteria. Eight medical databases were searched for studies published from Jan 1, 1997, to Jan 31, 2020. Original clinical studies in English assessing the association of factors detected during the febrile phase with progression to severe dengue were selected and assessed by three reviewers, with discrepancies resolved by consensus. Meta-analyses were done using random-effects models to estimate pooled effect sizes. Only predictors reported in at least four studies were included in the meta-analyses. Heterogeneity was assessed using the Cochrane Q and I² statistics, and publication bias was assessed by Egger's test. We did subgroup analyses of studies with children and adults. The study is registered with PROSPERO, CRD42018093363.Findings Of 6643 studies identified, 150 articles were included in the systematic review, and 122 articles comprising 25 potential predictors were included in the meta-analyses. Female patients had a higher risk of severe dengue than male patients in the main analysis ( 2674[16•2%] of 16 481 vs 3052 [10•5%] of 29 142; odds ratio [OR] 1•13 [95% CI 1•01-1•26) but not in the subgroup analysis of studies with children. Pre-existing comorbidities associated with severe disease were diabetes (135 [31•3%] of 431 with vs 868 [16•0%] of 5421 without; crude OR 4•38 [2•58-7•43]), hypertension (240 [35•0%] of 685 vs 763 [20•6%] of 3695; 2•19 [1•36-3•53]), renal disease (44 [45•8%] of 96 vs 271 [16•0%] of 1690; 4•67 [2•21-9•88]), and cardiovascular disease (nine [23•1%] of 39 vs 155 [8•6%] of 1793; 2•79 [1•04-7•50]). Clinical features during the febrile phase associated with progression to severe disease were vomiting (329 [13•5%] of 2432 with vs 258 [6•8%] of 3797 without; 2•25 [1•87-2•71]), abdominal pain and tenderness (321 [17•7%] of 1814 vs 435 [8•1%] of 5357; 1•92 [1•35-2•74]), spontaneous or mucosal bleeding (147 [17•9%] of 822 vs 676 [10•8%] of 6235; 1•57 [1•13-2•19]), and the presence of clinical fluid accumulation (40 [42•1%] of 95 vs 212 [14•9%] of 1425; 4•61 [2•29-9•26]). During the first 4 days of illness, platelet count was lower (standardised mean difference -0•34 [95% CI -0•54 to -0•15]), serum albumin was lower (-0•5 [-0•86 to -0•15]), and aminotransferase concentrations were higher (aspartate aminotransferase [AST] 1•06 [0•54 to 1•57] and alanine aminotransferase [ALT] 0•73 [0•36 to 1•09]) among individuals who progressed to severe d...