Statistics

TO THE EDITOR: We commend Heldens and colleagues1 for publishing their data on the incidence and in‐hospital mortality of sepsis and septic shock among patients admitted to Australian intensive care units (ICUs). The incidence of sepsis and septic shock in ICUs is estimated to be 101.8 and 19.3 per 100 000 patient‐years, respectively, at an attributable cost of $32 421.2 We concur that sepsis cases captured using the Australian and New Zealand Intensive Care Society Centre for Outcome and Resource Evaluation database criteria, compared with prospective clinical diagnoses,3 has poor sensitivity for sepsis case ascertainment. Notwithstanding, we propose that the application of a third surveillance metric using coded discharge data could be a viable alternative for sepsis case ascertainment and monitoring in ICUs. International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD‐10‐AM) diagnostic coding data are feasible to collect with a reduced risk of sampling bias and minimal loss to follow‐up. Using tandem dataset comparison following the implementation of a hospital‐wide sepsis pathway,4 we explored the utility of coding data for sepsis surveillance. We noted that 78% and 74% of ICU cases were designated an ICD‐10‐AM code denoting sepsis at admission and patient level, respectively (Box). Alarmingly, the concordance rate between coded administrative data and clinically verified sepsis diagnoses was even lower in non‐ICU settings. These data are in keeping with international reports.2 Robust and reproducible data are required to evaluate quality improvement regarding sepsis management. Given the poor sensitivity of research criteria and coding data, used in isolation for sepsis identification, a multifaceted approach is required. We hypothesise that the combination of administrative coding data and electronic medical record data, augmented with sepsis screening algorithms, may improve the sensitivity for sepsis case ascertainment in both cancer and non‐cancer settings.5 We encourage Heldens and colleagues to consider these suggestions as an alternative reproducible method needed to elucidate the incidence of sepsis and septic shock in Australian ICUs. Box – Relationship between sepsis cases satisfying clinical criteria and designated coded discharge data in intensive care unit (ICU) and non‐ICU settings, 2012–2014 Year Admission level ICU Non‐ICU All new admissions* ICD‐10‐AM captured cases Concordance All new admissions* ICD‐10‐AM captured cases Concordance 2012 38 27 71% 70 62 89% 2013 39 34 87% 175 103 59% 2014 81 61 75% 331 149 45% Mean (± SD) – – 78% (± 8.3%) – – 64% (± 22%) ICD‐10‐AM = International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification; SD = standard deviation. * Clinically diagnosed sepsis cases according to consensus diagnostic criteria.◆

Jake C Valentine · Gabrielle Haeusler · Leon Worth · Karin Thursky

The known: Recent advances in digital infrastructure in Australia allow linkage of administrative and clinical datasets.

Monique F Kilkenny · Joosup Kim · Nadine E Andrew · Vijaya Sundararajan · Amanda G Thrift · Judith M Katzenellenbogen · Felicity Flack · Melina Gattellari · James H Boyd · Phil Anderson · Natasha Lannin · Mark Sipthorp · Ying Chen · Trisha Johnston · Craig S Anderson · Sandy Middleton · Geoffrey A Donnan · Dominique A Cadilhac

Handbook of community-based participatory research

Fran Baum

To the Editor:Legislation mandates that all adult medical records be retained for a minimum of 7 years from the time of last patient contact, after which they can be destroyed. Exceptions to this requirement exist for young patients, and there are state-by-state variations, but there is no legislative requirement to retain records of patients who have implantable, high-risk devices. This is disturbing because many of these devices have an in vivo lifespan that exceeds 7 years. Of particular concern are patients with breast implants whose records may have been destroyed before a diagnosis of breast implant-associated anaplastic large cell lymphoma, which has an average latency period from implant to diagnosis of 9 years.1 Later presentations of this lymphoma are not uncommon, with latency intervals up to 23 years;2 therefore, it is imperative that implant details are retained to enable us to better understand the pathophysiology of this potentially fatal disease, which has been strongly associated with deeply textured surface implants. While the Australian Breast Device Registry (ABDR) is a safe repository for secure information on patients who are registered, those patients who are not may be at risk of losing important information about their implants. Furthermore, the expected lifespan of in vivo breast implants is at least a decade,3 so records may have been discarded at the time of patients presenting with serious implant-related problems. In our efforts to improve the safety of patients with breast implants, 30% of whom are breast reconstruction cases for cancer or congenital deformities, we encourage all practitioners to ensure that their patients are registered with the ABDR so their implant details are securely stored.4 In an effort to preserve the details of all Australian patients with breast implants, the ABDR can also store patient implant details retrospectively and will accept information from Australian patients having cosmetic tourism surgery overseas, after which significant complications can arise.5 It may be time, however, for legislation to be enacted to lengthen the mandatory retention period for patients with high-risk devices or to make it legally compulsory for practitioners inserting high-risk devices to enrol all patients into a clinical quality registry such as the ABDR.

Rodney D Cooter · Ingrid Hopper · John J McNeil

To the Editor:Ahern and colleagues1 explore the potential benefits and pitfalls of benchmarked reporting in the Australian context. As a binational registry of patients on renal replacement therapy in Australia and New Zealand, the Australia and New Zealand Dialysis and Transplant Registry has been producing and distributing centre-specific performance reports to renal units for over 20 years; these share many of the challenges faced by clinician-level reporting. In the past few years, this has extended to provision of an abridged version of the report on our website, containing unit-specific risk-adjusted outcome data for each dialysis and transplant unit (http://www.anzdata.org.au/v1/hospitalreport.html). The authors highlight the challenges of low case numbers resulting in statistical models that are underpowered to detect poor performance and require long observation periods that will limit timely detection of outliers and opportunities for remedial action. Co-opting statistical techniques used for quality control in other industries may present an opportunity to address these issues in the health care sector. Cumulative sum control charts2 provide a method for sequentially monitoring cumulative performance over time, which may permit early detection of poor performance and account for varying activity levels by including the number of procedures performed, rather than just a fixed time frame. Similarly, Bayesian approaches that involve updating prior probability distributions within a dynamic model may address these concerns3 and offer the conceptual advantage of explicitly testing not just the statistical difference from average but the likelihood of performance falling into a defined poor performance category. Finally, there are systems that use differing criteria for smaller and larger units.4 Ahern and colleagues discuss the potential consequences of poor performance, but omit any reference to exactly who should oversee this process. We assert that the relevant specialty or subspecialty body has a crucial role in overseeing the interpretation of reports. The detection of an outlier is dependent on the nature of the boundaries set for acceptable performance, and the vulnerability of the statistical adjustment model to bias and unmeasured confounders. Such interpretation requires detailed knowledge of the relevant field, an appreciation of the variation between centres, and substantial epidemiological knowledge.

Matthew P Sypek · Matthew D Jose · Stephen P McDonald