Identifying older diabetic patients at risk of poor glycemic control

Optimal glycemic control has been proved to prevent the onset and/or to slow the progression of several complications of diabetes and, thus, to improve quality of life [1–4]. Guidelines have been proposed to optimize the management of outpatients with diabetes and to assess the quality of the achieved glycemic control [5–16]. Attention has not been paid to the quality of glycemic control in people admitted to acute care hospital. Indeed, glycemic levels during the hospital stay, i. e. during a few days, are expected to lack any effect on the risk of complications which is related to the quality of glycemic control through several years. However, diabetic in-patients might represent a convenient sample for verifying whether any relationship exists between admission clinical/laboratory characteristics and the quality of glycemic control achieved at discharge and one year later. Targeting patients at risk of poor long term glycemic control could help promoting strategies to improve the quality of diabetes care.

Older diabetics represent the ideal population to be targeted because logistic problems frequently limit their access to ambulatory health care facilities. Furthermore, standards of care have been reported to be quite poor for a considerable proportion of older Medicare diabetic patients [17].

The present study has two objectives: 1) to develop a clinical prediction rule targeting older patients at risk of being discharged in poor glycemic control; 2) to verify whether the same prognostic model can predict long term poor glycemic control, as expressed by HbA1c > 7% one year later.

In an attempt to interpret the relationship between lower diastolic pressure and poor metabolic control, we characterized hypotensive patients and found that a greater Charlson's index (2.8 ± 2.2 vs. 2.5 ± 1.7, p < 0.005) and lower cholesterol level (169 ± 50 vs. 193 ± 53 mg/dl, p < .001) were the hallmark of this subset.

A score predicting poor glycemic control was developed as follows: each independent correlate of the outcome was assigned a score whose magnitude was directly proportional to the corresponding odds ratio. The score for admission glycemia > 218 mg/dl was arbitrarily established to be 5, while scores of the remaining predictors were computed according to the ratio between odds ratios of glycemia > 218 mg/dl and that of individual predictors obtained by the logistic regression analysis and were approximated to the unit. For example, the score of glycemia = 143–218 was computed by solving the following proportion: 2.84 : 5.29 = x : 5 and was found to be 2.7, which was approximated to 3. Insulin therapy prior to admission and diastolic arterial pressure inferior to 80 mmHg scored 2, Charlson's index > 2 scored 1. For each patient a final score was obtained by summing scores of individual predictors. Thus, a subject who used insulin prior to admission, had admission glycemia = 268 mg/dl and Charlson's index = 3 scored 8, i. e. 2 + 5 + 1 = 8.

In the testing population, HbA1c levels on admission and the predictive score were highly correlated (rho = 0.58, p < 0.001). Quartiles of HbA1c could predict the outcome glycemic control at discharge at least as effectively as the original and the modified predictive scores: HbA1c > 7.8 %, corresponding to the lower limit of the third quartile, had sensitivity of 85% and specificity of 60% versus the outcome poor glycemic control at discharge.

When compared with the predictive score and HbA1c >7.8% on admission, the modified predictive score showed the best mix of sensitivity (72%) and specificity (71%) towards the outcome HbA1c > 7% one year after discharge (Table 5). It could target accurately diabetics at risk of long term poor metabolic control (positive predictive value: 79%), but not those likely to have their diabetes well compensated one year after the discharge (negative predictive value: 58%).

	Score (level 4)	HbA1c >7.8% at admission	Modified score (level 5)
Sensitivity, %	66	69	72
Specificity, %	71	56	71
Positive predictive value, %	81	75	79
Negative predictive value, %	57	56	58

Sensitivity: True positives/(True positives + False negatives); Specificity: True negatives/(True negatives + false positives); Positive predictive value: True positives/(True positives + False positives); Negative predictive value: True negatives/(True negatives + False negatives).

This study demonstrates that over 50% of older diabetic patients are discharged from wards of general medicine and geriatrics of the acute care hospital in poor metabolic control and can be reliably identified by an admission value of HbA1c > 7.8 % or a simple clinical prediction rule. More importantly, these predictive tools effectively target patients exposed to a significant risk of poor long term glycemic control. The clinical prediction rule has greater specificity than HbA1c > 7.8% in predicting poor long term glycemic control, i.e. it outweighs HbA1c > 7.8% in targeting subjects who will have their diabetes well controlled one year after being discharged.

Patients with more advanced diabetes, as reflected by the systematic use of insulin at home and/or glycemia>218 mg/dl on admission, were at special risk of being discharged in poor glycemic control. Hypotension on admission was another risk factor for poor glycemic control likely because of its association with polipathology. Indeed, physicians are expected to encounter greater difficulty in tailoring the hypoglycemic therapy to the needs of patients in more unstable conditions and to compensate them more cautiously.

The analysis limited to the 1995–1997 period excluded that the mechanism of the payment based upon the Diagnosis-Related Group could have unduly shortened the stay and, thus, prevented the achievement of optimal glycemic control. On the contrary, in the United States the introduction of the Diagnosis-Related Group payment system could have worsened the quality of care provided to diabetic patients [31]. These observations caution against generalizing the present results to other health systems.

Achieving a satisfactory metabolic control is the only measure proven to prevent diabetic microvascular and neuropathic complications [1–4]. Unfortunately, such an objective was missed in a consistent proportion of diabetic patients, as 65.8% of them had high HbA1c levels one year after discharge. The necessity of balancing the benefit of glycemic control towards the risk of hypoglycemia only to some extent justifies such a result. Indeed, the protective effect of optimal glycemic control towards selected diabetic complications and the ensuing improvement in quality of life largely outweigh the risk of hypoglycemic episodes [1–4]. Furthermore, sustained improvement in glycemic control results in significant health cost saving with a lag time of 1 or 2 years [32]. However, our population was older than that enrolled in the UKPDS trials and by Wagner et al. Older age is associated with reduced sensitivity to neurovegetative effects of hypoglycemia [33]. Furthermore, social factors such as lack of both formal and informal support caution against an overzealous treatment of diabetes, mainly in patients with cognitive impairment or defective vision. However, lack of data on older diabetic populations prevented us from choosing levels of optimal metabolic control different from those reported to protect from microvascular complications. In this perspective, HbA1c > 7% likely represents a convenient index of poor long term metabolic control, whereas glucose ≥ 140 mg/dl seems a questionable short term index of metabolic decompensation. Indeed, achieving the optimal glycemic control might not be the primary outcome of in-patient management if comorbidity and frailty coexist with diabetes.

The recent experience shows that a considerable lag time exists between the unequivocal demonstration and the increase in the use of important and even lifesaving drugs such as ace-inhibitors for congestive heart failure; nevertheless, rate of use still remains far from the optimal one [34, 35]. It is likely that the optimization of hypoglycemic therapy will follow a similar time course. Indeed, the benefit of optimal glycemic control has been definitively proven in 1998 and further confirmed in 1999–2000. However, a pessimistic attitude towards diabetes rather than an insufficient knowledge of this disease seems to hinder the implementation of management guidelines by physicians, at least by primary care providers [36]. Furthermore the poor quality of documentation sent to the general practitioners after the patient's discharge from the hospital could contribute to worsen the subsequent management of diabetes [37].

Selected interventions such as a cooperative strategy between general practitioners and diabetologists or the organization of mini clinics by general practitioners or education programs can significantly improve metabolic control [38–41]. While these interventions are generically devoted to the average home-dwelling diabetic patients, present findings suggest that special attention should be paid to selected older patients discharged from the acute care hospital. These patients should benefit from a very careful supervision by the primary care providers and/or from programs aimed at optimizing the home management.

Limitations of our study deserve to be cited: 1) Fasting glycemia was the only measure of short term glycemic control. A glycemic profile would have provided a more complete and reliable estimate of the quality of therapy. However, morning hyperglycemia has been reported to be an uncommon consequence of nocturnal hypoglycemia in elderly diabetics on various therapeutic regimens [42]. Thus, it is likely that only a minority of patients classified as undertreated were exposed to an excess of hypoglycemic therapy at night. 2) Conclusions apply to older hospitalized patients and not to the general population. Thus, the proposed method aims at selecting diabetic patients amenable to special care programs after the discharge and, thus, optimizing the use of the few available economic resources. It has not been designed to target elderly homedwelling diabetics at risk of poor metabolic control; 3) The score was validated in a single General Medicine ward and not on multicenter basis. Thus, we cannot assume that discharge and therapeutic behaviors reflect those characterizing most of Geriatrics and General Medicine wards. This might limit the generalizability of results ; 4) Patients were not systematically screened for diabetic complications. This prevented us from verifying whether the presence of complications affected the therapeutic approach or had prognostic implications.

The present study shows that most of the older patients admitted to an acute care hospital with a primary or second-listed diagnosis of diabetes mellitus are discharged in poor metabolic control and continue to have high HbA1c levels one year after the discharge. Patients at risk of poor long term metabolic control can be effectively targeted by a simple clinical prediction rule on admission. However, attempts should be made to improve the diagnostic accuracy of the proposed clinical prediction rule by including selected presently untested variables such as the presence and severity of diabetic complications. Finally, research is needed to verify whether the identified predictors can be conveniently used to select patients amenable to dedicated home care programs.