First author [ref.] | Statistical method | Follow-up years | Outcome of the prediction model | Derivation/ validation | Procedure for variable selection | Initial predictors n | Final predictors n | Measure of association | AUC (95%CI) | HL Chi-squared | HL p-value | R2 | log2 likelihood |
Parshall [47] | Correlation analysis | up to 1 | Exacerbation (symptom-based) (healthcare use also analysed) | Derivation | Predefined variables# | 1 | 0.65–0.70# | 0.355–0.974# | 0.043–0.081# | ||||
Marin [41] | Correlation analysis | 5.1 (median) | Outpatient-treated exacerbation (hospitalisation also analysed) | Derivation | Predefined index | 1 | OR | 0.78 (0.73–0.84) | 0.31 | ||||
Jones [38] | Correlation analysis | up to 9 | Hospital admission for exacerbation | Derivation | Predefined index | 1 | 0.755¶ | ||||||
Chen [31] | Logistic regression | 14 days+ | Hospital readmission for exacerbation | Derivation | Predefined variables | 9 | OR | 0.10 | 139.49 | ||||
Almagro [27] | Logistic regression | up to 1 | Hospital readmission for exacerbation | Derivation | Univariable selection | 6 | 3 | OR | 1.9964 | 0.8496 | 0.146–0.194§ | ||
Brusse-Keizer [30] | Logistic regression | up to 1 | Patient with ≥2 exacerbations | Derivationƒ | Univariable and stepwise selection | 15 | 2 | OR | 0.717 (0.595–0.839) | 7.512 | 0.483 | 0.099–0.15§ | |
Faganello [33] | Logistic regression | up to 1 | Exacerbation | Derivation | Predefined index | 1 | OR | 0.69## | |||||
Suetomo [48] | Logistic regression | up to 1 | Exacerbation (hospital readmission for exacerbation also analysed) | Derivation | Predefined variables | 3 | 3 | OR | 0.77 | ||||
Takahashi [49] | Logistic regression | up to 1 | Patient with ≥2 exacerbations | Derivation | Predefined variables | 11 | RR | 0.81 | 66.64 | 0.56 | 0.22 | ||
Hurst [36] | Logistic regression | up to 1 | Increased exacerbation frequency | Derivation | Stepwise selection | 31 | 5 | OR | 0.22 | ||||
Bertens [29] (derivation) | Logistic regression | up to 2 | Patients with ≥1 exacerbation | Derivation | Backward stepwise selection | 6 | 4 | OR | 0.75 (0.69–0.82) | 8.66 | 0.37 | 0.26 | |
Bertens [29] (validation) | Not expected | up to 2 | Patients with ≥1 exacerbation | Validation | Not expected | Not expected | Not expected | Not expected | 0.66 (0.62–0.71) | 32.98 | 0.00 | 0.09 | |
Motegi [43] | Logistic regression | up to 2 | Exacerbation | Derivation | p-value selection | 14 | 2 | OR | 0.76–0.78++ | 0.15–0.18++ | |||
Thomsen [50] | Logistic regression | up to 1§§ | Frequent exacerbator (≥2 exacerbations >1 year apart) | Derivation | Predefined variables | 8–11ƒƒ | OR | 0.71–0.73ƒƒ | |||||
Lee [40] | Logistic regression | up to 0.5 | Any exacerbation (moderate–severe exacerbation (logistic regression) and time to first exacerbation (Cox regression) also analysed) | Derivation | Predefined variables | 1–9### | RR | 0.64–0.79### | |||||
Almagro [23] (derivation; CODEX) | Cox regression | up to 1¶¶¶ | Hospital readmission for exacerbation | Derivation | Predefined index | 1 | HR | 0.583 | |||||
Almagro [23] (derivation; ADO) | Cox regression | up to 1¶¶¶ | Hospital readmission for exacerbation | Derivation+++ | Predefined index | 1 | HR | 0.533 | |||||
Almagro [23] (derivation; BODEX) | Cox regression | up to 1¶¶¶ | Hospital readmission for exacerbation | Derivation+++ | Predefined index | 1 | HR | 0.633 | |||||
Almagro [23] (validation; CODEX) | Not expected | up to 1 | Hospital readmission for exacerbation | Validation | Not expected | Not expected | Not expected | Not expected | 0.590 | ||||
Echave-Sustaeta [32] | Cox regression | up to 1 | Hospital readmission for exacerbation | Derivation | Univariable and stepwise selection | 6 | 2 | HR | 0.7601 | 0.334 | |||
Gudmundsson [35] | Cox regression | up to 1 | Hospital readmission for exacerbation | Derivation | Predefined variables | Unclear | 5 | HR | 2175.6 | ||||
Garcia-Aymerich [34] | Cox regression | 1.1 (mean) | Hospital readmission for exacerbation | Derivation | Univariable selection | 23 | 7 | HR | 0.71 | ||||
Jakob [37] | Cox regression | 1.5 (mean) | Any exacerbation (outpatient-treated and inpatient treated exacerbations also analysed) | Derivation | Predefined variables | 7 | RR | 833.372§§§ | |||||
Müllerova [45] | Cox regression | up to 3 | Hospital admission for exacerbation | Derivation | Predefined variables | Unclear | 6 | HR | 0.742 (0.718–0.766)ƒƒƒ | 5.57ƒƒƒ | 0.696ƒƒƒ | 0.15ƒƒƒ | |
Ko [39] | Cox regression | up to 3 | Hospital readmission for exacerbation | Derivation | Predefined index | 1 | HR | 0.58#### | |||||
Moberg [42] | Cox regression | 5.5 (mean) | Hospital admission for exacerbation | Derivation | Univariable selection | Unclear | 18 | HR | 0.62 (0.59–0.65) | ||||
Ong [46] | Poisson regression | 1.35 (mean) | Hospital Admission for exacerbation | Derivation | Predefined index | 1 | IRR | 0.16¶¶¶¶ | |||||
Moy [44] | Negative binomial regression | 1.25 (mean) | Exacerbation (hospital readmission also analysed) | Derivation | Predefined variables | 4–5++++ | Rate ratio | 0.59–0.62++++ | |||||
Amalakuhan [28] | Random forest | up to 1 | Hospital readmission for exacerbation | Derivation§§§§ | Variables of importance evaluated with AUC | 60 | 5 | 0.75ƒƒƒƒ |
In this table only the data and analyses related to one outcome (in general the one deemed to be closer to the definition of exacerbation) are reported, with exception of [23], for which three different rows (for the three different predefined indices: CODEX (comorbidity, obstruction, dyspnoea and severe exacerbations), ADO (age, dyspnoea and airflow obstruction) and BODEX (body mass index (BMI), airflow obstruction, dyspnoea and severe exacerbations)) are used. Where data are not displayed, they were not reported and/or not straightforward to evaluate. The number of initial predictors is often a proxy we have evaluated. AUC: area under the curve; HL: Hosmer-Lemeshow test; OR: odds ratios; RR: relative risk; HR: hazard ratios; IRR: incidence rate ratios. #: four-domain subscales of the Medical Outcomes Study short-form health survey (36 items) (general health perceptions, mental health, role limitation-physical, correlated mental-health score) were evaluated as separated prediction models; ¶: performance refers to the DOSE (dyspnoea, obstruction, smoking and exacerbation frequency) index (a performance was also discussed for Medical Research Council Dyspnoea Scale and forced expiratory volume in 1 s % pred); +: follow-up of 3 months also analysed; §: Cox-Snell-Nagelkerke definition of R2; ƒ: to assess over-fitting jackknife cross validation technique was applied to the prediction rule; ##: performance refers to the BODE (BMI, airflow obstruction, dyspnoea and exercise capacity) index (a performance for Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage is also provided) and other multivariable models are also shown but no performance was provided; ++: according to the model used (four different ones are presented); §§: the supplementary material of the study reports the data for a median follow-up of 4 years (range of performances with and without the three biomarkers evaluated in the study: 0.92–0.92); ƒƒ: according to the inclusion in the final model of three inflammatory biomarkers as predictors, net reclassification index used to evaluate (40%; 22–57%) the improvement in performances with the inclusion of the three biomarkers; ###: according to if chronic obstructive pulmonary disease (COPD) assessment test was categorised or not and if there was or not adjustment for other variables (age, BMI, duration of COPD, current smoking status, number of comorbidities, history of influenza vaccination, country); ¶¶¶: for the derivation cohort the follow-up of 3 months was also analysed; +++: validation was performed but no performance was provided; §§§: Akaike information and Schwarz-Bayes criteria also provided, but not reported here; ƒƒƒ: data provided upon request by the authors; ####: performance refers to the BODE index (a performance for the BODE's 4-component was also provided); ¶¶¶¶: performance refers to the BODE index (a performance for GOLD stage was also provided); ++++: different models evaluated (Step_CRP, STEP_IL-6 or BODE); §§§§: internal validation performed; ƒƒƒƒ: mean AUC for 200 runs.