Non-recruitment to and selection bias in studies using echocardiography in haemodialysis patients

作者：Chiu Diana Y Y; Green Darren; Abidin Nik; Hughes John; Odudu Aghogho; Sinha Smeeta; Kalra Philip A^*

来源：Nephrology, 2017, 22(11): 864-871.

DOI：10.1111/nep.12865

• dialysis
• exclusion
• mortality
• recruitment

BackgroundIt is unknown whether patients recruited to renal cardiac imaging studies are fully representative of the investigated population and whether there are differences in characteristics and survival between participants and non-participants (excluded or declined consent). Subjects and Methods Four hundred thirty-five maintenance haemodialysis patients were screened in an observational, prospective study. Baseline demographics, laboratory results, social deprivation scores and survival data were collected from patient records. All patients were followed-up until death, renal transplantation or 16 November 2015.
ResultsForty-four patients were excluded (16 language barrier, 10 mental incapacity, 9 severe co-morbid illness and 9 because of immobility), 172 patients declined consent (84% due to reluctance to attend for an extra visit) and 219 patients were recruited. Excluded patients had a lower mean haemoglobin (10.2g/dL vs 10.7g/dL), phosphate (4.15mg/dL vs 4.74mg/dL), albumin (3.6g/dL vs 3.8g/dL) and higher C-reactive protein (3.2mg/dL vs 1.6mg/dL) compared with recruited patients. No difference was identified between groups for Charleston comorbidity index (P=0.115) or social deprivation scores. After a median follow-up of 29.7 (25th-75th percentile, 21.1-34.3) months, there were 141 deaths. In a multivariable Cox regression model adjusting for BMI, age, Charleston comorbidity index, haemoglobin, albumin, smoking status and diabetes mellitus, patients who declined consent had an adjusted HR of 1.70, 95% CI 1.10-2.52, and excluded patients had an adjusted HR of 1.30, 95% CI 0.75-2.25, for all-cause mortality compared with recruited patients.
ConclusionsPatients recruited to the study had longer survival compared with non-participants. Research studies should document phenotypes of non-participants to aid interpretation and generalizability of results.
Summary at a Glance The article is an excellent piece of work. It is clear, concise, and avoids hiding the true intent under excessive data or result-reporting, or over-analysis. The true message is not regarding echo outcomes but rather, it tells a clear and much needed cautionary tale regarding how we so easily trust cherry-picked data that may be far from real-world reality.

• 出版日期2017-11