Table 5.
Summary of findings table
| Which is the diagnostic accuracy of validated algorithms based on health administrative data to diagnose heart failure compared to clinical diagnosis? | |||||
| Population | General adult population including subgroups based on demographics or comorbidities | ||||
| Prior testing | Some studies preselected subjects with a simpler version of the algorithm | ||||
| Setting | Primary care, outpatients or inpatients | ||||
| Index test | Case-detection algorithms from routinely collected health data | ||||
| Importance | Algorithms based on administrative health data are valuable to detect large cohort with heart failure rapidly and inexpensively | ||||
| Reference standard | Clinical diagnosis performed by a clinician or health professional (medical examination or medical chart review) | ||||
| Studies | 14 cross-sectional studies, 8 longitudinal studies (using multiples contacts with the health system over 1–3 years), and 2 case–control studies | ||||
| Subgroup | Accuracy (95% CI) | No. of participants (studies, algorithms) |
Prevalence in the sample used for validation of the algorithm Median (range) |
Practical implication | Quality and comments |
| Outpatient or primary care general population |
Range • Sensitivity 24.8% (95% CI 22.3–27.5%) to 97.3% (95% CI 97.0–97.6%) • Specificity 35.6% (95% CI 30.4–40.8%) to 99.5% (95% CI 99.4–99.6%) • No pooled analysis due to heterogeneity |
103,018 (6, 14) | 5.6 (2.9–9.8) |
The estimated prevalence of the disease is between 1 and 4% and 10% over 70 years; the HF prevalence of the studies (case control excluded) is in this expected range The PPV ranged from 22.4 to 84.5% |
For the patient selection domain, two studies had high risk of bias; one of them also applicability concerns. Another study had high risk of bias in the flow and timing domain. Poor reporting issues were found, especially concerning patient selection. One study did not report the number of diseased |
| Subgroup | Accuracy (95% CI) | No. of participants (studies, algorithms) |
Prevalence Median (range) |
Practical implication | Quality and comments |
| Hospitalized patients |
Range • Sensitivity 29.0% (95% CI 27.4–30.6) to 96.0% (95% CI 91.0–99.0) • Specificity 84.3% (95% CI 84.0–84.6) to 99.2% (95% CI 98.8–99.6) • No pooled analysis due to heterogeneity |
14,957 (8, 10) | 13.7 (9.3–100) |
There are not reliable estimates of the prevalence in unselected hospitalized population, which are however expected to be higher than in the outpatient setting, as is the case in the analyzed studies The PPV ranged from 35.8 to 94.0% |
Six out of eight studies had quality concerns: for the patient selection domain, two studies had high and one study an unknown risk of bias, and an additional study had applicability concerns. Four studies had high risk of bias in the reference standard domain. Five studies had high risk of bias in the flow and timing domain |