Background

Interactions between diseases, nutritional status and medical treatment become complicated with advancing age. Often, old age is accompanied by multimorbidity as well as increased vulnerability to drug adverse effects [1, 2]. These may result in nutritional imbalance, including malnutrition. Malnutrition is as an important dimension of the elder care quality [3] and is a major cause of vulnerability to stress (frailty) [4].

Malnutrition risk may be monitored by measuring metabolic indexes. Notable examples include Body Mass Index (BMI), and potentially - glycated hemoglobin (HbA1c%) and total serum cholesterol (herein referred to as cholesterol). Many current quality improvement programs monitor metabolic indexes (e.g. [5, 6]). Avoiding high values is the prevalent target while low values are seldom flagged. However, since the correlation of these metabolic indexes with mortality yields a U-shaped curve (see below), low values of these indexes might need some more attention.

Rapid weight loss, especially with low BMI values, usually mandates investigation in order to eliminate the cause or to limit its effect [7, 8]. In a meta-analysis that included 2.88 million people, the optimal BMI for overall survival was found to be 25–30 kg/m2 [9] (with a U-shaped curve). However, an optimal individual BMI in the elderly population is a debated issue [10,11,12,13]. High-quality nutritional-intervention trials were reported as few and too small in size (26–210 patients each) [14]. These trials used different BMI cut-offs (18.5–24 kg/m2) and different weight-loss cut-offs (2.5–10%) as inclusion criteria. Outcomes other than weight change were reported in even fewer trials, which showed trends towards improvements at best. Hence, more research is needed to draw conclusions whether older adults should receive nutritional counseling to avoid weight loss.

The optimal HbA1c% values for survival were 7.5–8% and mortality increased under the 6.5% level among patients who take two antidiabetic medicines [15]. This suggests an increased metabolic deficit risk in tight anti-diabetic phamacological treatment. Since a substantial proportion of older adults with diabetes is potentially overtreated [16], an intervention to reduce overtreatment could be benefitial.

Correlation between death and cholesterol is also U-shaped [17]. Hypocholesterolemia, defined as cholesterol below 160 mg%, predicts increased mortality and morbidity [8, 18]. Clinical questions about statin overtreatment, especially in primary prevention for the elderly, are still unanswered [19]. Prevalence of hyperlipidemia overtreatment, defined as statin treatment of low-risk patients (less than 5% 10-year cardiovascular risk based on the Framingham Heart Study equation) was estimated at 8% [20].

Hence, these low metabolic indexes could potentially be markers for patients that may benefit from appropriate intervention to attenuate the indexes’ decrease. We hypothesized that informational intervention via e-mail to primary physicians and nurses could positively affect the health of these patients. Unfortunately, the scientific evidence regarding e-mail use for clinical communication between healthcare professionals is sparse [21]. While use of computer reminders in family medicine has been used for some decades to effect physician actions (e.g. [22]), we did not find scientific literature on interventions in the community that are similar to our trial (i.e. using e-mail, addressing low metabolic values). However, a hospital based automatic e-mail alert system was found useful in effectively screening patients at risk of malnutrition [23]. Our trial (LIMIT) is thus aimed at investigating the effects of intervening in these identified high risk groups by way of sending one-time e-mail reminders to the primary physician and nurse about their patient’s low metabolic values.

The objectives of LIMIT are shown in Table 1. All outcome measures will be collected from the national CHS computerized database. The primary objective is to determine if the intervention improves patient survival across the groups of high risk patients. The key secondary objectives are to determine if the intervention influences staff response and patient morbidity. Staff response will be assessed by comparing rates of nurse and dietitian evaluations and de-prescribing. Morbidity will be assessed by comparing rates of clinic and emergency-room visits, costs, total drug use, hypoglycemia events etc. The number needed to treat (i.e. number of e-mails sent) to prevent any death in one year will be assessed as a measure of clinical significance [24].

Table 1 Outcomes, variables, measures and methods of analysis
Full size table

Methods/design

LIMIT is a randomized, controlled, open label, superiority trial with 1:1 allocation of two parallel groups. LIMIT is conducted in the CHS community clinics of the Northern and Southern Districts, where CHS service most of the population (71%, 62% respectively), and an even higher percentage within elderly and vulnerable populations. With a 2-year BMI recording rate of 73.2% in our population, a 99.8% rate of recording HbA1c% in patients taking diabetes drugs, and a 99.1% rate of cholesterol level recording, LIMIT is a unique ‘real life’ trial of community medicine. The schedule is presented in Table 2.

Table 2 LIMIT schedule of enrolment, intervention and assessment
Full size table

A pre-trial letter was sent to the primary physicians and nurses in order to introduce the trial protocol and to provide a way to allow them to refrain from participating. Data was extracted from the CHS computerized medical database. Inclusion and exclusion criteria are presented in Table 3.

Table 3 LIMIT inclusion and exclusion criteria
Full size table

Since some participants fit more than one inclusion criterion, 7 subgroups were created, presented in Table 4.

Table 4 LIMIT subgroups
Full size table

Randomization was generated by an excel computation (‘RAND’ function) that was repeated automatically until equal numbers of participants (≤ 2% differenceFootnote 1) were achieved in all subgroups. The procedure was stopped automatically by using the ‘Goal Seek’ function. The first author activated and recorded the result of the randomization process.

The one-time intervention letter provided relevant patient data and an alert to the primary care providers (physician and nurse) about low values of BMI, HbA1c% or cholesterol with an advice to consider appropriate dietary and medical revision. The intervention e-mails were created by using Microsoft Word ‘Mail Merge’ feature. (Examples are shown in the trial protocol in the Additional file 1). These e-mails were sent automatically using Microsoft Outlook with the ‘Request a Read Receipt’ option checked. Emails were resent, during the first 3 months, only if a “recipient’s mailbox is full” message was received. Open discussions between mail recipients and researchers were encouraged by all means to ensure safety and to add efficacy. All data files and e-mails are stored in secure CHS servers. All trial investigators will have access to the data. The authors adhere to the SPIRIT guidelines for the reporting of trial protocols. The study results will be released to the participating districts’ personnel and to the general medical community. Authorship policy will follow the recommendations of the International Committee of Medical Journal Editors. Regrettably, open data sharing cannot yet be guaranteed.

The intervention arm (patient-specific reminder e-mail letter) will be compared with the control (standard care) for all primary and secondary analyses (Table 1). The proportion of deaths between two groups will be also compared by using a logistic regression model in order to control for confounders: age, gender, BMI, HbA1c%, Cholesterol level, previous MI, IHD, CVA, TIA. Since the primary outcome is death from any cause, we based study power calculation on previously observed mortality differences (1.8% vs. 3%) after e-mailing similar reminders about sulphonyl-urea treatment. Thus, study power of 80% (alpha = 0.01, two-tail Pearson chi-squared test) is achieved by studying 3906 people in each arm. Recruiting 2 CHS districts was needed to reach this number.

Categorical variables will be shown as frequencies and percentages. Continuous variables will be shown using standard distribution indices (e.g. average, standard deviation, median, etc.). Differences between the arms of the study will be examined using Chi-square test (or Fishers’ exact test) for categorical variables, T-test for continuous variables with normal distribution and nonparametric Wilcoxon two sample test for continuous variables without normal distribution. Missing data will be handled using ‘Rubin’s rules’ of multiple imputation and details of the sensitivity analyses will be provided. The mediation of survival differences by the secondary outcomes will be assessed by examining correlation of primary outcome versus secondary outcomes at the subject level, e.g. by performing a logistic regression. The statistical processing will be performed using Excel or SAS 9.2 software and will be statistically significant if P < .05 (or P < .01 for the primary outcome).

Descriptive statistics are shown in Table 5. Data (up-to-date to 30.9.15) of all 75 years and older members: 26,491 in the Southern district and 22,132 in the Northern district, was collected. For LIMIT’s flow diagram, see Additional file 2.

Table 5 Baseline characteristics of the population and of LIMIT participants
Full size table

BMI < 23 kg/m2 was found in 6159 patients (12.7%). BMI drop of at least 2 kg/m2 in 2-years was found in 4051 patients (8.3%). Dietitian counseling was reported for 1159 patients (2.4%) during previous year, and only for 38 of the 867 patients who had BMI < 23 kg/m2 after losing 2 kg/m2. Criterion ‘a’ was met by 845 patients (1.7%).

Diabetes drugs were given to 12,131 patients (25%), of whom (4186, 8.6%) had HbA1c% ≤ 6.5% (criterion b).

Cholesterol lowering drugs were given to 24,013 patients (49.4%). 10,232 of them (21%) had hypocholesterolemia. After exclusion of patients with ischemic cardiac or cerebral diseases, 4423 (9.1%) were included for criterion c.

More than one criterion was met by 862 patients (1.8%). A total of 8584 (17.7%) patients were included and were randomized to intervention or control. Of the included participants, 4977 (58%) were women and 488 (5.7%) were aged 90 years or more.

After randomization, 4310 patient-specific alerts for intervention were prepared by using Mail Merge feature in Microsoft Word. These emails were sent between 2 and 9 November 2015, to 506 physicians and 155 nurses at 383 clinics. Up to 10.12.15 (one month later), 2233 (52%) reading confirmations were received. While sending the intervention e-mails to the physicians and to the Northern district nurses, a concern was raised by some recipients regarding their workload. Hence, the researchers decided to avoid sending the e-mails to the Southern district nurses.

Discussion

LIMIT was jointly developed by a multi-professional primary-care research team that included family-physicians, geriatricians, nurses and a dietitian. The model for the intervention derived from the currently implemented use of e-mails to alert primary care staff about patient specific ‘quality measures’ focusing on inadequately high indices. A trial addressing malnutrition and drug-overtreatment as new ‘quality measures’ seemed worthwhile.

Regarding the prevalence of LIMIT criteria, a direct comparison with published data is difficult. About 15%–20% elderly patients experience a loss of either 5 kg or more or 5% of usual body weight over 5–10 years and the incidence of unintentional weight loss in studies involving adults seeking health care varies from 1.3% to 8%, depending on the setting and definition of weight loss [25]. Thus, our data may fit within the higher estimations, possibly due to older age. We found no data to compare the rate of dietitian counseling in community weight-losing patients.

We found a similar prevalence of potential overtreatment of diabetes (criterion b) in people aged 65+ in US adults [16]. The baseline prevalence of hypocholesterolemia in our cohort (36% overall and 22% in patients treated with cholesterol-lowering drugs) is higher than estimation based on Lipid Research Clinic Data, 1983 [26] and by Elmehdawi [27] – where hypocholesterolemia prevalence was estimated at 2% to 6% of the elderly. This may reflect differences in cutoff levels and in age of participants (older age comes with wider diversity of measures), a global surge in use of cholesterol-lowering drugs [28] and other reasons.

The high prevalence of elderly people who fit the inclusion criteria, underscores LIMIT’s potential public importance.