The study was carried out at the wards of the Department of General Internal Medicine at the University Hospital MAS, in Malmö, Sweden. Data was collected from October 2009 through June 2010. At the time, the hospital was a 700-bed teaching hospital, providing healthcare to the community of Malmö and its surrounding areas. The hospital is now part of the larger Skåne University Hospital.
The Department of General Internal Medicine consisted of four wards with a similar general medical profile. The average length of stay was 6.4 days. Most patients (90-95%) were admitted through the hospital’s Emergency Department (ED), the rest were directly referred via their GPs.
The study included only the hospital’s general internal medicine wards; all medical departments with a higher degree of specialisation (Endocrinology, Angiology, Haematology, Nephrology, Gastroenterology, Rheumatology, Cardiology, the Department of Infectious Diseases and the Department of Pulmonary Diseases) were excluded. At the general internal medicine wards, patients tended to be older or to have multiple comorbidities to a larger extent than at the more specialised wards. The hospital’s bed manager, unaware of our study, designated the patients to their wards.
Whether at the ED or directly at the wards, a staff physician examined all patients on arrival, documenting presenting complaint, past medical history, drug history and examination findings. A nurse also assessed all patients when arriving at the wards. This included standardised estimates of the risk of falls and pressure sores, using the Downton Fall Risk Index and the Modified Norton Scale, respectively [23, 24]. These procedures were done according to hospital policy.
Eligible patients were 60 years or older, residing in the city of Malmö and not living in a nursing home. Occasionally, eligible patients were not available to enter the study, e.g. if patients were put in isolation due to norovirus infection.
Study personnel approached eligible and available patients on the first or second day of their stay to determine appropriateness for cognitive testing. For example, patients with terminal disease or severe aphasia were considered inappropriate. Patients were considered inappropriate at admission if a possibly reversible condition was present, such as severe delirium (incoherent speech, inability to focus attention) and/or abnormal laboratory values (Haemoglobin < 100 g/L, temperature > 38°C, C-Reactive Protein > 50 mg/L, abnormal electrolytes). These patients were assessed continuously and, if the condition resolved, they were subsequently included.
Eligible, available and appropriate patients were approached regarding consent. All included patients gave their written informed consent. If cognitive tests disclosed significant cognitive impairment, written consent from an informant was collected as well. This procedure was approved by the regional ethics committee at Lund University.
Three experienced research assistants (two certified occupational therapists and one registered nurse) carried out the measurements at the ward, in a private and calm environment, between 8 am and 4 pm.
Interviews and comorbidity
Interviews were held with patients concerning living situation, family, education and access to home care. Presenting complaints noted in the charts on admission were recorded. Past medical history was extracted from the hospital’s charts, all conditions noted during the current or three preceding admissions were recorded. Frequent conditions were classified as absent/present. The list of current medications in the medical records was examined on admission and the cumulative number of drugs was noted. We used Charlson comorbidity index to obtain a standardised estimate of comorbidity .
The MMSE, (Mini-Mental State Examination) and the CDT (Clock-Drawing Test) were employed [26, 27]. The MMSE is scored from 0–30, with 0 points representing maximum cognitive impairment. As a cut-off, ≤ 23 points was used. In the CDT, the patients were asked to draw a clock on a sheet of paper and add the hands of the clock showing ”ten past eleven”. The CDT was scored from 0 to 5 according to Shulman, where 0 points denote maximum cognitive impairment and 5 points a perfect clock . Any scoring uncertainties were discussed within the group until a consensus was reached. As a cut-off, ≤ 3 points were used. Previous studies have described a correlation between MMSE and CDT scores, with a mean correlation coefficient of 0.61, and that combining the tests gives higher diagnostic accuracy for neurocognitive disorders . The CDT is also less affected by depression than the MMSE .
The Quality of Life in Alzheimer’s Disease (QoL-AD) scale was employed . The QoL-AD contains 13 items (Physical health, Energy, Mood, Living situation, Memory, Family, Marriage, Friends, Self as a whole, Ability to do chores around the house, Ability to do things for fun, Money and Life as a whole). The items are rated from 1 to 4 where 1 represents poor, 2 fair, 3 good and 4 excellent. The rating can be done by patients and/or proxies.
Subjective memory complaints (SMCs)
SMCs were determined in two ways. Firstly, a direct yes/no question was asked: ‘Do you think that your memory has gotten worse lately?’ Secondly, the memory item of the QoL-AD scale was used. This item was dichotomised, with a score of 1–2 denoting subjective memory complaints and 3–4 no impairment.
If available at the hospital, informants rated the patients’ quality of life with the QoL-AD scale. As for the patients, the memory item was dichotomised and a score of 1 or 2 denoted cognitive impairment. This was done separately from the patients’ ratings.
Recognition by staff physicians
The staff physicians’ admission notes from the first day were reviewed. Any notation of neurocognitive disorders (dementia, delirium or MCI – mild cognitive impairment) or current symptoms (disorientation, memory impairment, confusion, irrational behaviour etc.) was considered as recognition of cognitive impairment.
Recognition by staff nurses
The nurses’ admission notes were examined. The Downton Fall Risk Index comprises a yes/no item entitled ”cognitive impairment” . The Modified Norton Scale includes the item ”mental condition”, that is scored from 1–4 (with 1 representing ”no contact”, 2 ”cannot answer adequately”, 3 ”occasionally confused” and 4 ”fully oriented”) . Scores other than ”no cognitive impairment” or ”fully oriented” on any of the scales was considered recognition of cognitive impairment.
The included patients were also taking part in a prospective intervention study aiming to reduce hospital readmissions. Interventions included a pharmacist’s medication review and a changed discharge routine. For the present study, only intervention status (control or intervention) was recorded to rule out the possibility of this confounding the results.
Patients were divided into three groups according to their results on cognitive tests, with 0, 1 or 2 abnormal test results, where patients with 1 abnormal result could have a low score either on the MMSE or on the CDT. All other baseline variables were compared between the three groups. In the primary comparison, ANOVA and chi-square tests were used where appropriate. In secondary analyses we did pairwise comparisons between groups with Bonferroni correction for multiple comparisons. The correlation between MMSE and CDT scores was determined using Spearman’s rho.
Bivariate Cox proportional hazards regressions were done separately for demographic variables, comorbidity variables and cognitive tests, adjusting for age and sex where applicable. The ‘number of abnormal cognitive tests’ variable was dummy-coded to compare 1 abnormal test vs 0 and 2 abnormal tests vs 0. The assumption of proportional hazards (that relative risk is not time-dependent) was tested using log-log plots and time-interaction tests, no violations of the assumption were found.
For the multivariable analysis a stepwise approach was carried out, using a backwards method with p >0.051 as the threshold for removal. Starting the stepwise model with all variables or only the ones with a bivariate p value of <0.25 resulted in the same final model. Exclusion of categorical variables with small cells (neurocognitive disorder) did not affect the final model. Testing for multiple collinearity revealed a moderate correlation between ‘Charlson comorbidity index’ and ‘number of drugs’ (Spearman’s rho, r = .46), exclusion of the latter did not affect the final model. For each step our model was controlled and fit the data adequately.
All calculations were done using the SPSS software (SPSS version 19.0, SPSS inc. Chicago Illinois). A two-sided p value of < 0.05 was considered significant.